Computational and Experimental Techniques Towards Optimising the Cardiovascular Risk Assessment of Hyperbaric Decompression Stre

Imperial College London

Department of Bioengineering

Computational and experimental techniques towards optimising the cardiovascular risk assessment of hyperbaric decompression stress caused by circulatory bubble dynamics

Virginie Papadopoulou

Submitted in partial fulfilment of the requirements for the degree of

Doctor of Philosophy of Imperial College London.

Declaration of Originality

I hereby certify that this thesis is a product of my own work. Where I have consulted the published work of others, this has been appropriately referenced. Where other people have contributed to the work presented, this has been clearly stated and their contributions attributed.

Virginie Papadopoulou.

The copyright of this thesis rests with the author and is made available under a Creative Commons Attribution Non-Commercial No Derivatives licence. Researchers are free to copy, distribute or transmit the thesis on the condition that they attribute it, that they do not use it for commercial purposes and that they do not alter, transform or build upon it. For any reuse or redistribution, researchers must make clear to others the licence terms of this work.

Virginie Papadopoulou.

Abstract

This work focuses on developing new techniques towards the quantification of hyperbaric decompression stress. Instead of just preventing decompression sickness (DCS), the aim is to go towards developing an environmental cardiovascular personalised stress index, especially as sub-clinical long terms effects of even recreational scuba diving have been demonstrated. From an engineering perspective, despite the longevity of the research field, a number of fundamental issues that remain unknown have prevented efficient modelling. The aim of this thesis is to directly tackle the research methodology by developing three tailored tools.

Firstly, we develop a simulation platform in MatLab to model the diving process by optimizing the implementation of dissolved gas phase tracking decompression algorithms. This platform can be used to simulate diving scenarios, but also analyse real dive profiles. From a first analysis on real profiles provided to us by the European Divers Alert Network database, we find as expected that these existing models are poor predictors of accidents, but also demonstrate that ascent rate seems to be an important predictor of DCS for the range of profiles considered.

Secondly, a fundamental issue for modelling the decompression phenomenon is that the precise formation site and growth mechanism of decompression bubbles in vivo remains unknown. We develop a novel experimental set-up and analysis code for the real-time optical study of decompression induced bubble growth dynamics. Looking at bubble growth from a gas saturated solution on ex-vivo muscle and fat tissues, we show that the role of the substrate from which bubble detach plays a significant role. Bubble density, nucleation threshold, detachment size and coalescence behaviour are shown significantly different for the two substrates, whereas growth rates after a critical size are governed by diffusion as expected, and a competition for dissolved gas between adjacent multiple bubbles is demonstrated. These findings are not accounted for in current modelling efforts so our experimental set-up could be used in the development of a more physiologically relevant decompression model.

Thirdly, an important question in terms of decompression modelling optimisation is the precise definition of the evaluation endpoint. Vascular circulating bubbles are normally assessed semi-quantitatively by trained human raters who grade the severity on echocardiograms. We show statistically that this is highly rater-dependent compared to a new 7 counting methodology which is found to perform significantly better but is more time- consuming. We then use image processing techniques to semi-automate this new counting methodology with good comparison to human raters, significantly reducing the time needed for the assessment. This new method could be added to decompression model validation protocols, as well as used in physiology experiments looking at predictive parameters for, or preventive measures against, circulating gas bubbles post-dive.

The proposed experimental and computational techniques could be used towards optimising the cardiovascular risk assessment of hyperbaric decompression stress caused by circulatory bubble dynamics.

Acknowledgements

“From birth, man carries the weight of gravity on his shoulders. He is bolted to earth. But man has only to sink beneath the surface and he is free”

—Jacques Yves Cousteau, Oceanographer and Explorer

When I first started looking into decompression modelling for scuba diving as a one year research project, little did I know I would end up embarking on a PhD. As a physicist and keen scuba diver myself, I had become fascinated by understanding why modelling efforts had not been more successful in the past. On starting this project, I quickly realised the physiology associated to scuba diving was much more than just pressure effects and encompassed cold exposure, exercise, immersion, stress, etc. Within months of diving into the literature and starting my own research I grew increasingly convinced I would want to do this for a longer period of time and started searching how to fund a full PhD. Now after three years on this topic, I believe I have made some small new contributions to the field that will prove useful in the endeavour towards the quantification of a personalised stress index associated to scuba diving. The last three years have been for me a learning journey on many levels, with a new passion discovered and incredible opportunities along the way.

I should thank my diving instructor George Filios for conveying his passion for the sport with me as a child and training me all the way into the first levels of professional recreational diving. Interestingly, he intuitively implements a lot of the physiological advice we are only now discovering through scientific studies. Without him, I would never have discovered this field.

I am eternally grateful to both Dr Mengxing Tang and Dr Robert Eckersley who welcomed me into their research group when they really had no reason to. I have learnt a great deal about contrast enhanced ultrasound imaging in the process. I am also indebted to Prof. Costantino Balestra and Prof. Thodoris Karapantsios for agreeing to co-supervise and add their expertise to this collaborative endeavour. I am much obliged to Prof. Balestra for funding me through the PHYPODE project when a position became available in his team and 9 giving me an invaluable perspective of human research and physiology training. I am also very thankful to Prof. Karapantsios for numerous in depth discussions and for hosting me for three months in his laboratory to perform experiments on bubble growth. I could not have dreamed of a better supervisory team and I am extremely lucky that they were not only excellent supervisors but also excellent mentors. It is interesting how all four were complementary in the training and scientific advice they offered me through their expertise in different fields, yet invariably consistent both in reinforcing good scientific practice and in their mentoring advice.

I would like to thank our Ultrasound Imaging Group at Imperial College London and King’s College London, including the students I supervised during my PhD, in particular Joe How Hui and Chris Song, as well as Prof. David Cosgrove, Dr James Choi, Dr Jennifer Siggers, Prof. Kim Parker and Prof. Eleanor Stride for useful feedback. Thank you to all the PHYPODE supervisors and fellows for brainstorming future directions in the field and to the Multiphase Dynamics Group of the Aristotle University of Thessaloniki for their warm welcome and hosting. Finally, I am very lucky to have had the chance to interact with other established researchers in the community who have all been very approachable and thank in particular Dr Thodoris Mesimeris, Dr Adel Taher, Mr Jean-Pierre Imbert, Prof. Saul Goldman, Prof. Radek Pudil, Dr Antoine Boutros, Dr Ole Hyldegaard, Dr Andreas Mollerlokken, Prof. David Doolette and Prof. Ran Arieli.

I am also very grateful to DAN Europe and in particular Massimo Pieri, Danilo Cialoni and Dr Alessandro Marroni for generously giving us access to their diving research database used in Chapter 3.2.

With respect to Chapter 4, I owe to acknowledge in more detail the help of: Dr Sotiris Evgenidis in setting up the experimental unit; Christos Ampatzidis for pH and conductivity analyses; Rania Oikonomidou for helping out in the last experiments; Dr Kelly Pavlidou and Prof. Ioannis Savvas of the Veterinary School for the handling of the animals and tissue conservation advice; Dr Mesimeris for generously providing his small pressure chamber to modify and use; and Dr Thodoris Mesimeris and Prof. Margaritis Kostoglou for fruitful discussions on the direction of this work.

I would also like to thank Dr Peter Germonpré, Dr George Obeid, as well as Walter Hemelryck for their help for Chapter 5: they were instrumental in explaining the technique

10 and challenges of acquiring echocardiography post-dive in the field before I had the chance to participate in field experiments. Without their previous work in optimising this process from an acquisition point-of-view, much of the quantification effort presented here would not have been nearly as successful.

Furthermore, I should acknowledge the funding sources that have supported the research findings presented here: the PHYPODE project, financed by the European Union under a Marie Curie Initial Training Network Program (EU-FP7-ITN-264816), the European Network Action COST MP1106 “Smart and green interfaces – from single bubbles and drops to industrial, environmental and biomedical applications”, the Imperial College (IC) Trust and European Underwater and Baromedical Society (EUBS) travel grant support, as well as the Imperial College (IC) library open publication fund.

Finally, I thank my family and in particular my father, brother, grand-mother and husband Nikos, for their continued love and support through all the ups and downs of my PhD journey.

Dedicated to my father

Contents

Abstract ...... 7

Acknowledgements ...... 9

List of Figures ...... 21

List of Tables ...... 25

Chapter 0 – Thesis structure and publications ...... 27

Chapter 1 – Background ...... 31

1.1 Context and definitions: scuba diving ...... 31 1.1.1 Historical perspective ...... 31 1.1.2 Current diving classifications ...... 33 1.1.3 Decompression illness ...... 35

1.2 Decompression modelling ...... 38 1.2.1 Dive computer validation ...... 38 1.2.1.1 Defining “validation” ...... 39 1.2.1.2 Operational considerations ...... 40 1.2.1.3 Decompression calculations ...... 40 1.2.1.4 Proposed lifecycle for dive computers ...... 42 1.2.2 Brief historical overview of decompression models ...... 42 1.2.2.1 Dissolved gas phase models: Haldanean-based decompression theory ...... 42 1.2.2.2 Silent bubbles and deep stops: redefining DCS ...... 44 1.2.2.3 Tracking free-phase gas (bubbles): dual phase models ...... 47 1.2.2.4 Alternatives and challenges to modelling assumptions ...... 48 1.2.3 Conclusion ...... 50

Chapter 2 – Literature review and PhD aims ...... 53

2.1 A critical review of bubble formation in hyperbaric decompression ...... 53 2.1.1 Aim and scope...... 53 2.1.2 Background ...... 54 2.1.2.1 Fundamental physics ...... 54 2.1.2.2 Early Studies ...... 55 2.1.2.3 The micronuclei stability problem ...... 58 2.1.3 Recent Studies and Discussion ...... 59 2.1.3.1 Bubble formation mechanisms ...... 59 2.1.3.2 Stability...... 60 15

2.1.3.3 Single and multiple bubble behaviour ...... 62 2.1.3.4 Role in decompression modelling...... 64 2.1.4 Conclusion ...... 67

2.2 Circulatory bubble dynamics: from physical to biological aspects ...... 69 2.2.1 Aim and scope...... 69 2.2.2 Background: how can bubbles end up in the blood stream? ...... 70 2.2.2.1 Bubbles introduced in the bloodstream on purpose ...... 71 2.2.2.2 Unwanted bubbles in the bloodstream ...... 73 2.2.2.3 Life cycle of a bubble in blood ...... 75 2.2.3 Bubble growth and detachment from decompression ...... 77 2.2.3.1 General formalism: heat and mass transfer ...... 77 2.2.3.2 Thermal degassing (heat transfer controlled) ...... 78 2.2.3.3 Decompression degassing (mass transfer controlled) ...... 79 2.2.3.3.1. Pool degassing (diffusion controlled) ...... 79 2.2.3.3.2. Flow degassing (inertia controlled) ...... 81 2.2.3.4 Bubble detachment ...... 81 2.2.3.4.1. In stagnant liquid ...... 81 2.2.3.4.2. In flowing liquid ...... 82 2.2.4 Bubble behaviour in the bloodstream ...... 85 2.2.4.1 Bubble dissolution in blood ...... 85 2.2.4.2 Bubble dynamics in an ultrasound field ...... 87 2.2.4.3 Rheology of microbubbles in the bloodstream ...... 88 2.2.4.3.1. Brief overview of blood rheology ...... 88 2.2.4.3.2. Interfacial tension and surfactants ...... 90 2.2.4.4 Biological interactions ...... 91 2.2.5 Conclusion ...... 92

2.3 Rationale behind PhD research ...... 94 2.3.1The need for hyperbaric decompression stress quantification ...... 94 2.3.2 Defining PhD aims ...... 95

Chapter 3 – Development of the Modelling Framework ...... 97

3.1 Aim and Scope ...... 97

3.2 Development of a MatLab decompression platform implementing dissolved gas phase tracking algorithms ...... 97 3.2.1 Simulation platform inputs and outputs ...... 97 3.2.2 Basic code description ...... 99 3.2.2.1 Principles ...... 99 3.2.2.2 Individual Functions ...... 102 3.2.3 Decompression profiles...... 104 16

3.2.3.1 Example result ...... 104 3.2.3.2 Corresponding compartment analysis ...... 105 3.2.4 Comparison against known tables ...... 106 3.2.4.1 DSAT No-Decompression Limits ...... 106 3.2.4.2 US Navy decompression tables ...... 108

3.3 Analysis of real dive profiles (dissolved gas phase) ...... 109 3.3.1 Methods ...... 110 3.3.1.1 Harmonization of data format entry...... 110 3.3.1.2 Modifications and validation for analysis of real dive profiles...... 110 3.3.1.3 Statistical analyses ...... 113 3.3.2 Results ...... 114 3.3.2.1 Epidemiological findings ...... 114 3.3.2.2 Relationship between accidents and degree of conservatism ...... 115 3.3.3 Discussion ...... 116 3.3.4 Conclusion ...... 117

Chapter 4 – Development of Novel Experimental Set-up ...... 119

4.1 Aim and Scope ...... 119

4.2 Development of experimental set-up and analysis code ...... 120 4.2.1 Schematic of full experimental set-up developed ...... 120 4.2.2 Development of novel experimental set-up ...... 123 4.2.2.1 Chamber modifications and testing ...... 123 4.2.2.2 Saturated liquid and inlet into/ outlet out of chamber ...... 123 4.2.2.3 Temperature control: sensors, heating belts and PID ...... 124 4.2.2.4 Design of inside container and emptying mechanism ...... 127 4.2.2.5 Optical data acquisition set up and lighting ...... 128 4.2.3 Experimental procedure ...... 129 4.2.3.1 Detailed Steps ...... 129 4.2.3.2 Camera acquisition configuration for bubble growth and density ...... 130 4.2.4 Image processing analysis code in MatLab...... 132 4.2.4.1 Circular Hough Transform for bubble recognition (CHT) ...... 134 4.2.4.2 Robustness additions (DMP) ...... 135 4.2.4.3 GUI implementation for semi-automatisation ...... 135 4.2.5 System Evaluation ...... 136

4.3 Experiment and analysis fat/muscle ...... 137 4.3.1 Material and methods ...... 137 4.3.1.1 Experimental Procedure ...... 137 4.3.1.2 Theoretical analysis for bubble growth ...... 137 4.3.1.3 Data Fitting and Statistical Methods ...... 140 17

4.3.2 Results ...... 141 4.3.2.1 Bubble Density ...... 142 4.3.2.2 Growth Rate ...... 142 4.3.2.3 Detachment Size and Delay Times in Cyclic Growth ...... 144 4.3.2.4 Contact Lines and Contact Angles...... 144 4.3.2.5 Multiple Bubbles: competition for dissolved gas and coalescence ...... 145 4.3.4 Discussion ...... 146 4.3.5 Conclusions ...... 149

Chapter 5 – Development of new endpoint evaluation using ultrasound imaging ...... 151

5.1 New bubble counting methodology statistical comparison ...... 151 5.1.1 Introduction ...... 151 5.1.2 Methods ...... 154 5.1.2.1 Statistical methods ...... 158 5.1.3 Results ...... 162 5.1.4 Discussion ...... 164 5.1.5 Conclusions ...... 167

5.2 Use on real dive data from field experiments ...... 169 5.2.1 Attersee experiments ...... 169 5.2.1.1 Background ...... 169 5.2.1.2 Results and Discussion ...... 170 Trimix dives ...... 170 Air dives ...... 171 5.2.2 NEMO33 experiments ...... 171 5.2.2.1 Aims and constraints ...... 171 5.2.2.2 Experimental plan ...... 172 Additional considerations ...... 174 5.2.2.3 Bubble counting results ...... 175 5.2.2.3 Discussion...... 177 5.2.3 Conclusion ...... 178

5.3 Automatizing the counting of VGE on echocardiograms ...... 180 5.3.1 Introduction ...... 180 5.3.2 Methods ...... 181 5.3.2.1 Data and statistical comparison ...... 181 5.3.2.2 Brief overview of the image processing steps ...... 182 Step A: ROI segmentation of the right heart chambers ...... 183 Step B: Selection of open valves frames for bubble counting ...... 183 5.3.2.3 Additional processing steps details ...... 184 Segmentation of the right heart chambers (step A) ...... 184 18

Selection of open valves frames (step B) ...... 185 Bubble counting ...... 186 5.3.3 Results ...... 188 5.3.3.1 Variation assessment and agreement with manual raters ...... 188 5.3.3.2 Time spent ...... 188 5.3.3.3 Semi-automatizing of the thresholds used ...... 189 5.3.4 Discussion ...... 192 5.3.3 Conclusion ...... 194

Chapter 6 – Conclusion ...... 195

6.1 Summary of contributions ...... 195

6.2 Discussion of limitations and future work ...... 197

6.3 Final thoughts ...... 198

Appendix A - PUBLICATIONS ...... 199

Appendix B - CV ...... 205

References ...... 209

List of Figures

Figure 1 : A 1846 diagram of a caisson devised by Jules Triger ...... 31 Figure 2 : Photograph of the Eads Bridge where 42 workers developed DCS ...... 32 Figure 3 : Typical recreational scuba diving gear [Wikimedia Commons by Soljaguar] ...... 34 Figure 4 : Richard Pyle puncturing the bladder of a fish while ascending from the depth...... 45 Figure 5 : Supersaturation of fast and slow compartments for a shallow and deep stop profile. A shows the two 30min air decompression profiles tested; B the supersaturation of a fast (halftime=10min) compartment for these two profiles; and C the supersaturation of a slow (halftime 160min) compartment for these two profiles...... 46 Figure 6 : A horizontal diver during decompression ...... 50 Figure 7 : Ultrasound imaging of a human liver with microbubble contrast agents: non-linear mode preferentially showing bubbles (left) and standard linear B-mode image (right) ...... 72 Figure 8 : Diagram of targeted microbubbles shell composition [3] ...... 73 Figure 9 : Ultrasound imaging of the heart 1 hour post scuba dive. Venous gas emboli (VGE) are circulating in the right heart chambers (delimited in white, image reversed). The lungs are effectively filtering these VGE and they do not appear in the left heart chambers...... 75 Figure 10 : Schematic of bubbles from decompression in the circulation (inverted orientation in keeping with echocardiography frames) ...... 75 Figure 11 : Geometrical definitions of parameters for the velocities and force balance equations [211] ... 83 Figure 12: Input and output of decompression simulations plotted on a model decompression profile. .... 98 Figure 13: MatLab GUI interface for user input (breathing mixture, bottom time, maximum depth and calculation method) and output display...... 99 Figure 14: Simplified (ignoring ascent rate) graphical representation of the M-value concept, showing that the tolerated gas loading varies with depth, where point A is the start of the decompression. . 101 Figure 15: Dive at 40m for 20min breathing air throughout the dive...... 104 Figure 16: Partial pressure of inert gas for the 5 chosen compartments and corresponding dive depth during the decompression portion of the dive...... 105 Figure 17: Corresponding ascent ceiling. At the beginning of the decompression, the diver is allowed to ascend to 6m since the highest pressure ceiling is 1.53 ata (5.3m depth, rounded up to the biggest multiple of 3, therefore 6m)...... 106 Figure 18: Nitrogen partial pressure for the compartments for a theoretical profile of 40min to 40m with air calculated using the decompression platform in Section 3.2 from bottom time and depth input (least conservative ascent, in black). In colour: calculated with the analysis platform throughout the full dive ...... 111 Figure 19: Top: Depth (red) and worst conservatism ratio (blue) throughout the dive; Bottom: percentage of breathing gases throughout the dive, useful in case of technical diving with gas switching...... 113 Figure 20: Comparison of worst degree of conservatism (Student T-test) between DCS (1.28 ± 0.27) and no DCS (1.23 ± 0.14) groups; no significant statistical difference (p=0.456)...... 115

Figure 21: Comparison of fastest ascent rate (Mann-Whitney U-test) between DCS (12.0 ± 4.4 m/min) and no DCS (9.1 ± 3.7) groups; significant difference (p=0.0047) ...... 115 Figure 22: Photograph showing the optical acquisition system and the temperature controlled small pressure chamber, as well as an example result...... 121 Figure 23: Top view schematic of experimental set-up, showing liquid (in blue) and gas (in red) pressure flow systems...... 122 Figure 24 : Back view of the chamber before subsequent changes (but already fitted with new airtight connectors) ...... 123 Figure 25 : Fitted connectors for the tube (red here) to allow the liquid flow in and out ...... 124 Figure 26 : Front view of the chamber showing the tube that lets the saturated liquid in the chamber .. 124 Figure 27 : Liquid saturation tank fitted with heating belt for temperature control ...... 125 Figure 28 : Front (left) and back (right) views of the chamber once fitted with the purpose-built heating rings. All ring are plugged into the same multi-plug switched on and off by a PID device (temperature control)...... 125 Figure 29 : Temperature control overview of chamber set-up ...... 126 Figure 30 : Thermal sensor fitted to the chamber (airtight purpose-built seal): inside front view (left) and outside back chamber view (right) ...... 126 Figure 31 : Container and base, front view as it goes in the pressure chamber (left) and back view (right) showing the water inlet for the saturated liquid; also used to drain ...... 127 Figure 32 : Emptying container system ...... 128 Figure 33 : lighting set-up (chamber view from the front, open glass door removed)...... 128 Figure 34 : Optical acquisition set-up for bubble growth data (front view) ...... 129 Figure 35 : Automated acquisition control for picture acquisition every 5 seconds ...... 131 Figure 36 : Picture of a trial configuration for resolution testing ...... 131 Figure 37 : Camera set-up for density acquisition ...... 132 Figure 38 : Flowchart of analysis code in MatLab. CHT: circular Hough transform; DMP: decision making process to keep only one radius and centre per bubble if multiple choices given in CHT output, by taking into account the previous frame output for centres and/or ROI and/or radius value (depending on user choice for DMP input parameters)...... 133 Figure 39 : Example output from MatLab program, showing the result after circular Hough-transform (right, axes in pixels) and the output with the centre and radius of the bubble marked on the original image (left, axes in pixels)...... 134 Figure 40 : Example of bubble recognition with bubble outline drawn in blue and centre position displayed in red (axes in pixels) ...... 135 Figure 41 : Bubble density comparison between fat (n = 11) and muscle (n = 11) tissue substrates ...... 142 Figure 42 : Superimposed muscle (red) and fat (blue) radius (μm) versus time (s)...... 143 Figure 43 : Bubble radius at detachment (left, fat n = 8, muscle n = 7) and delay times between bubbles growing from same nucleation site (right, fat n = 7, muscle n = 6 and multiple n = 6), for muscle and fat tissue substrates...... 144 Figure 44 : Example bubble tracking result from custom MatLab code ...... 144 22

Figure 45 : Inverse relation between distance between two adjacent bubbles and their average G coefficient...... 148 Figure 46: Landmark structures in the right heart echography image: the upper circle identifies the ‘top’ of the right ventricle (RV) while the lower two circles identify the section through the tricuspid annulus on either side of the right atrium and constitute the ‘upper’ border of the RA. (N.B., echocardiograph images are inverted)...... 155 Figure 47: Choice of frame to analyse: the three landmark circles are drawn as in Figure 46. The frame chosen for analysis is indicated by the red marker on the electrocardiography trace (marked by the small green circle, bottom right). Both leaflets of the tricuspid valve are fully open and visible against the ventricular wall (points of green arrows); the right atrium and ventricle form a single cavity...... 156 Figure 48: Bubble counting: bubble signals are identified as bright spots and counted individually; tricuspid valve leaflets and other fixed structures (e.g., papillary muscles in the top of the right ventricle) are not counted...... 156 Figure 49: Bland-Altman plot showing systematic over-estimating by Cardiologist 3 as compared to the mean number of VGE counted by all others; X-axis: number of VGE in the image; Y-axis: difference of count vs. mean; horizontal lines – 95% CI as 1.96 std differences; LoA – limits of agreement ...... 163 Figure 50: Bland-Altman plot showing the good consistency between reference score and all observers for frame-based counting in the video sequences; X-axis: number of VGE in the video sequences (average of 10 frames); Y-axis: difference of count vs. mean; horizontal lines – 95% CI as 1.96 std differences; LoA – limits of agreement ...... 163 Figure 51: Bubble counting score for all 6 subjects after the Attersee air dives using the new counting methodology (as described in Section 5.1) ...... 171 Figure 52: Example schedule of pre and post dive measurements for one group during the NEMO experiments...... 174 Figure 53: Evolution of bubble score post dive, NEMO experiments (10 subjects) ...... 176 Figure 54: Example of echocardiography frame for two subjects, a) and b), at all four post dive measurement time points; subjects a and b correspond respectively to subjects 4 and 8 in Figure 53...... 177 Figure 55: Example frame of post-dive echocardiography showing VGE in the right heart chambers (right) and corresponding anatomy of the heart (left)...... 181 Figure 56: (a) Example frame ; (b) Septum identification from intensity boundary; (c) segmentation ROI result; (d) Identified bubbles within the ROI ...... 182 Figure 57: Overview of the image processing steps: a user-set intensity threshold (T) is used to segment the right heart chambers defined as the ROI; the area of the ROI evolution in time is then used to identify frames where the heart valves are open (maximum area) for bubble counting...... 183 Figure 58: Use of the connected component labelling for identification of the septum and muscle wall. In the case where the binarisation operation (example result shown in a) disconnects the two structures

(figure b) these are linked by a straight line (shown in c and d) to avoid counting structures in the arterial side and outside the cone shaped field of view...... 185 Figure 59: Plot of ROI area against frame number ...... 186 Figure 60: Corresponding sinusoidal fitting to Figure 59 ...... 186 Figure 61: Bubble counting steps shown for one frame a/ after threshold T for the whole cone region; b/ corresponding identified ROI; c/ Masking of steps a and b for bubble counting ...... 187 Figure 62: Corresponding original frame and final output with counted bubbles to Figure 61 ...... 187 Figure 63: Original frame and corresponding trichromatic coefficients ...... 189 Figure 64: Example of user segmentation used for threshold intensity calculation ...... 190 Figure 65: Histogram of the trichromatic green component, showing a characteristic bimodal distribution ...... 190 Figure 66: Example of 2 term Gaussian Mixture Model curve-fitted histogram for the trichromatic green component ...... 191 Figure 67: Comparison of the area of the ROI with respect to the frame index for one video. In blue: manual user segmentation and in red: semi-automatic segmentation area...... 191 Figure 68: Heart rhythm and ROI variation illustrated for a flex (stress) echocardiogram showing a region (black circle) with higher variation than for typical rest recordings...... 193

List of Tables

Table 1 : Comparison overview between circulatory bubble types found in the bloodstream...... 71 Table 2: Table with Nitrogen and Helium half-times (in min) used for the 16 compartments, as per the Bühlmann ZH-L16 algorithm [281] ...... 100 Table 3: No Decompression Limit (NDL) for depth exposure, calculated from the MatLab decompression platform and from the DSAT planner (PADI)...... 107 Table 4: Example decompression procedures for some square profile air dives using USN tables [50] and from the MatLab decompression platform. For example: A diver having stayed at 21.4m depth for 100min will need to make two decompression stops according to the MatLab platform (4min at 6m then 31min at 3m) and one stop according to the USN table (33min at 3m)...... 108 Table 5: Spencer grading system for VGE on echocardiograms post dive ...... 152 Table 6: Eftedal and Brubakk system for VGE grading post dive on echocardiograms ...... 153 Table 7: Proposed interpretation for Kappa Statistic values...... 160 Table 8: Static images bubble counting – identical image pairs scores; Spearman ρ between raters and a reference score (see text); all comparisons non-significant (Wilcoxon test-retest p>0.05); C – cardiologist, MD – physician, O – other (paramedic or hyperbaric chamber attendant) ...... 161 Table 9: Planned organisation of subject groups for each day, NEMO experiments ...... 173 Table 10: Rotation of measurement stations for every timepoint, NEMO experiments ...... 173 Table 11: Timepoints of measurement pre and post dive for the NEMO experiments; the times stated correspond to the start of Station A, then +7min for following stations ...... 173

Chapter 0 – Thesis structure and publications

The thesis is structured as follows:

Chapter 1: Background

Introduces scuba diving history and the evolution of decompression models for decompression sickness prevention.

Chapter 2: Literature review and PhD aims

Presents the literature related to bubble formation, growth and dynamics, and discusses these in the context of hyperbaric decompression in vivo. The conclusions are used to determine some of the main challenges in the field and propose PhD aims.

Chapter 3: Development of the Modelling Framework

Introduces a platform developed in MatLab to model the diving process by optimizing the implementation of dissolved gas phase tracking decompression algorithms. The platform is used to analyse real dive data from the Divers Alert Network database.

Chapter 4: Development of Novel Experimental Set-up

Proposes a novel experimental set-up and analysis code for the real-time optical study of decompression induced bubble growth dynamics. The set-up is used to analyse the differences in bubble behaviour between ex-vivo fat and muscle substrate.

Chapter 5: Development of new endpoint evaluation using ultrasound imaging

Compares a new counting methodology to current venous gas emboli assessment by trained raters and demonstrates it performs significantly better. This method is then semi-automated using image-processing techniques, significantly reducing the time needed for the assessment.

Chapter 6: Conclusion

Results are summarised and the impact and limitations of the contributions discussed. Possible future directions from this work are also examined. 27

Resulting1 Publications:

Refereed journal papers

 Papadopoulou V, Eckersley RJ, Balestra C, Karapantsios TD, Tang M-X. A critical review of physiological bubble formation in hyperbaric decompression. Advances in colloid and interface science. 2013;191–192(0):22-30. [1]

 Germonpré P, Papadopoulou V, Hemelryck W, Obeid G, Eckersley RJ, Tang M-X, Balestra C. The use of portable 2D echocardiography and 'frame-based' bubble counting as a tool to evaluate diving decompression stress. Diving and Hyperbaric Medicine; 2014 Mar;44(1):5-13. [2]

 Papadopoulou V, Tang M-X, Balestra C, Eckersley RJ, Karapantsios TD. Circulatory Bubble Dynamics: From Physical to Biological Aspects. Advances in colloid and interface science. 2014; 206:239-249. [3]

 Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Kostoglou M, Tang MX, Karapantsios TD. Decompression induced bubble dynamics on ex-vivo fat and muscle tissue surfaces with a new experimental set up. Colloids and Surfaces B: Biointerfaces. 2015; 129:121-129. [4]

Conference proceedings

 Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Tang M-X, Karapantsios, T. A study of decompression induced bubble dynamics on different tissue surfaces with a novel experimental set-up. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

1 Please refer to the Appendix for the complete list of publications including other collaborative work not directly related to this thesis. 28

 Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Tang M-X, Karapantsios, T. Decompression induced bubble growth on tissue surfaces. Smart and Green Interfaces Conference – COST MP1106, Marseille, 22-24 April 2014.

 Papadopoulou V, Hui J, Balestra C, Hemelryck W, Germonpré P, Eckersley R, Tang M. Evaluating the counting of venous gas emboli on post-scuba dive echocardiographs. Tri-continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013.

 Papadopoulou V, Evgenidis S, Eckersley RJ, Balestra C, Tang M-X, Karapantsios, T. Decompression induced bubble growth on tissue surfaces from gas saturated solutions. Tri-continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013.

 Papadopoulou V, Hui J, Balestra C, Hemelryck W, Germonpré P, Eckersley R, Tang M. Automated Counting of Venous Gas Emboli in Post-SCUBA Dive Echocardiography. IEEE-UFFC conference, Prague, Czech Republic, 21-25 July 2013.

 Papadopoulou V, Evgenidis S, Eckersley RJ, Balestra C, Tang M-X, Karapantsios, T. Effect of different tissue surfaces on decompression induced bubble growth from gas saturated solutions. COST MP1106 conference, Prague, Czech Republic, 20-24 March 2013.

 Papadopoulou V, Eckersley RJ, Balestra C, Tang M-X. Decompression modelling: a projection procedure for dissolved phase tracking simulations. EUBS 2012 conference, Belgrade, Serbia, 11-15 Sept 2012.

Chapter 1 – Background

1.1 Context and definitions: scuba diving

1.1.1 Historical perspective2

The first documented case of decompression sickness [5] was Robert Boyle's (1627-1691) experiments with a vacuum pump, which saw the appearance of a bubble in the eye of a viper he was experimenting on, after he decreased the pressure by mechanically removing air from its containing glass box: “I observed the viper furiously tortured in our exhauster receiver, which had manifestly a conspicuous bubble moving to and fro in the waterous humor of one of its eyes” [6].

Figure 1 : A 1846 diagram of a caisson devised by Jules Triger [Wikimedia Commons by Amédée Burat]

2 Adapted from V. Papadopoulou, Early Stage Assessment Report, Imperial College London, Bioengineering Department, 2012. 31

Decompression illness was rediscovered in the 19th century as “caissons disease” when diving bells (or “caissons”) started being used in underwater construction (Figure 1). These were notoriously used in the 1850s and 1860s in the construction of the Medway Bridge in Rochester, England and that of the Eads Bridge over the Mississippi river [7, 8] (Figure 2). The caisson workers complained of pains in the joints upon returning to the surface and the symptoms were so severe that a lot of men died or became permanently paralysed as a result. The bent posture adopted by caisson workers suffering from severe joint pain gave decompression illness its other name: “the bends”. However it was noticed that spending only a short amount of time underwater and returning slowly to the surface was a way of avoiding the problem and thus these guidelines were gradually introduced empirically, but yet with no physical explanation as to why they should make any difference.

Finally it was Paul Bert (1833-1886) who was able to explain why divers suffered from the bends in his work La Pression Barometrique [9], where he explained that pressure affects them chemically in changing the proportions of oxygen in the blood (thus explaining the Central Nervous System oxygen toxicity problem), and how the body absorbs inert gases (in this case nitrogen) which it then has to release gradually during ascent so that the bubbles formed can be exhaled: “for they must not only allow time for the nitrogen of the blood to escape but also to allow the nitrogen of the tissues to pass into the blood”. In addition, he was also the first to suggest what we now call “deep stops” (the reader is referred to Chapter 3), and to propose the use of pure oxygen to treat decompression illness.

Figure 2 : Photograph of the Eads Bridge where 42 workers developed DCS [Wikimedia Commons by Kbh3rd]

1.1.2 Current diving classifications2

Recreational SCUBA is an acronym that stands for 'Self Contained Underwater Breathing Apparatus'. As the name suggests, scuba diving is the activity of swimming underwater while carrying all the air necessary so as to avoid surfacing for a certain amount of time, nowadays typically around 45 minutes for recreational diving.

Scuba diving as a recreational activity grew more popular around the 1960s, as standardised equipment (Figure 3) and safe diving practices were adopted and enforced through internationally recognised training agencies such as the Professional Association of Diving Instructors (PADI), the British Sub-Aqua Club (BSAC) and French “Confédération Mondiale des Activités Subaquatiques” CMAS to name a few.

Recreational diving is limited to a maximum depth of 40m and does not include decompression stops by limiting the time spent at depth (bottom time). In this type of diving, the breathing gas is usually compressed air, brought to ambient pressure by the regulator, or Nitrox which is a mix of nitrogen and oxygen in different percentages. Nitrox is over- oxygenated air [4] (most common combinations Nitrox32 and Nitrox36 containing 32 and 36 percent oxygen respectively) and allows the diver longer bottom times since the amount of nitrogen is reduced. However care must be taken with regards to maximum depth as the partial pressure of oxygen cannot exceed 1.4 (as dictated by guidelines) to avoid any oxygen toxicity (above partial pressures of 1.6-1.8) so Nitrox is especially useful for “shallow” dives (34m and 29m maximum for Nitrox32 and Nitrox36 respectively).

Technical In technical diving, a more advanced recreational sport, the main difference is that saturation is now allowed and depth not limited anymore. The decompression then often needs stops at different depths for certain amounts of time. In addition divers carry more equipment and most often use Trimix instead of air, a mix of oxygen, nitrogen and helium, (or Heliox which contains oxygen and helium only) whose concentrations are dictated by the dive profile, and nitrox or even 100% oxygen in the last stages of decompression to maximise off-loading of nitrogen and helium and therefore cut decompression times [10].

Commercial Finally commercial diving should also be mentioned. It is professional diving to perform work underwater, for instance related to the oil industry (exploration, drilling, etc), civil engineering (underwater welding, bridge foundation building, supervision, etc). There are other types of professional diving, in archaeology, marine biology, etc, but these 33 are usually done using the same techniques as in technical diving protocols (more or less). The difference with what is defined here as “commercial” diving is that in most cases the divers are directly connected to the surface (or underwater pressurised chamber) and a compressor that provides them with air (“lifeline”) and thus do not carry a SCUBA in the sense of its acronym.

These dives are also very different from the recreational and even technical sectors in that the diver often stays “compressed” for days, weeks and even months, living in a pressurised cabin, until the work to be done is completed, and is then brought back to surface pressure only once (instead of doing it every day which would increase the risk of decompression sickness but also lose valuable time and money for the company by spending hours decompressing). This type of diving is called “saturation diving” because the body of the diver becomes saturated with inert gas, to the point where he is not on-gassing anymore so any additional time spent underwater does not add to his decompression after saturation. This decompression is usually done for the most part inside a decompression chamber either after the diver surfaces quickly and is recompressed, or from depth.

Figure 3 : Typical recreational scuba diving gear [Wikimedia Commons by Soljaguar]

1.1.3 Decompression illness3

Decompression Illness (DCI) is a pathophysiology affecting divers, astronauts, pilots and compressed air workers. It is caused by bubbles which grow in the body during or after a reduction in ambient pressure (decompression). DCI encompasses both arterial gas embolism (AGE) and decompression sickness (DCS) which can be difficult to distinguish and require the same initial treatment [11]. AGE which can also have iatrogenic causes results from gas emboli in the arterial circulation, either from a pulmonary over expansion which ruptures the alveolar capillaries or through cardiac shunts that allow venous gas emboli to enter the arterial circulation. DCS, also referred to colloquially as “the bends”, is caused by bubble formation from dissolved inert gas in the tissues during decompression.

In the case of scuba diving, pressurized air (or another breathing mixture) is breathed by the divers at ambient pressure throughout the dive. As pressure increases with depth the partial pressures of oxygen and inert gases breathed are also increased. This results in a pressure gradient from the inspired gas in the lungs to the rest of the tissues in the body which are saturated for sea level. As the divers descend and stay at depth, inert gases, not utilized by the body will dissolve in the tissues until these become saturated. The uptake of gas happens with different rates for different tissue types. Once the divers start to ascend, the pressure gradient reverses and the tissues start “off gassing” creating bubbles that go from the tissues into the blood stream. Normally these bubbles diffuse from the alveolar capillaries into the lungs to be expired out of the body through respiration. The mechanism by which bubbles formed in extravascular tissue can end up in the blood is unknown, but the lymphatic pathway has been proposed as a potential entry route into the circulation [12, 13]. Doppler Ultrasound findings have shown repeatedly that bubbles are formed routinely on dives [14-19] and only sometimes does it result in DCS. This can happen when the ascent is too fast for instance, yielding big bubbles which get stuck in a blood vessel and/or too many bubbles which overload the filtering capacity of the lungs. Another mechanism proposed is that of very

3 Adapted from V. Papadopoulou et al. Adv Colloid Interface Sci.,2013;191–192(0):22-30.

35 small bubbles passing through the lungs into the arterial circulation and being subsequently excited to growth by gas diffusion from nervous tissues [20].

There are an estimated 7 million active recreational scuba divers worldwide and the world's biggest training agency, PADI, certifies over 500,000 new divers every year [21], with annual certifications tripling in the last 20 years [22]. Additionally, diving is also key for environmental and scientific monitoring, construction and maintenance work, offshore oil exploitation, forensic, rescue, military and filming purposes. In the USA over 1100 cases of DCI are reported every year, 100 of which are fatal [21]. In the absence of complications relating to asthma, shunts and lung infections or diseases, AGE can be prevented effectively by adhering to slow ascents and to the golden rule of diving “never hold your breath”. Occurrences have decreased dramatically from 18% of total DCI occurrences in 1987 to 8% in 1997 [23]. In a study of DCI data from 1998 the Divers Alert Network (DAN) classified as AGE only 3.9% of 441 cases [24]. In contrast to AGE, DCS risk is inherently dependent on the dive profile and most importantly on the ascent profile. It is managed by adhering to decompression schedules dictated by tables or dive computers which allow for stops at different depths for controlled off gassing of tissues so that bubbles can be effectively eliminated by respiration. DCS occurrence is also relatively rare, with rates of 0.01–0.1% per dive, the higher end of the spectrum reflecting rates for commercial diving and the lower rates for scientific and recreational diving [25-28]. DAN's study on a population of 135,000 dives made well within the current limits for decompression by 9000 recreational divers showed a DCS rate of 0.03% [24, 29]. Some studies with different decompression procedures show significantly higher risk, for instance 1.3% for some US Navy dives in the 70s [30] or 4.4% in US Navy trials for long exposures under increased exercise and thermal stress [31]. The definition of acceptable risk also varies widely depending on the diving purpose, commercial diving setting it at 0.1% for mild and 0.025% for serious cases, and the US Navy at 2% for mild and 0.1% for serious cases [32].

A number of predisposing factors have been identified for DCS, hydration levels being one of the most important [33, 34]. A lot of studies have been done with regards to a potential link between a PFO (Patent Foramen Ovale) and an increased DCI risk [35, 36]. The idea of a systematic PFO screening for all divers is not implemented due to the prevalence of the condition in the general population (roughly one in four people) and the debatable benefits of having it surgically closed versus the risk of the operation itself [37]. The general consensus

36 remains however that in the event of a previous DCI case, then a PFO screening should be undertaken. Obesity, temperature, smoking, age, repetitive diving, flying after diving, reverse or toothpick diving profiles, as well as previous injuries are also nowadays considered to be risk factors [38, 39].

The role of exercise has also been debated [40] and depending on its timing and intensity can increase or decrease risk [41-44]. Additionally, an adaptive response to diving has been hypothesized and the susceptibility to DCS seems to be very different from individual to individual [45]. The bubbles can cause problems through mechanical effects directly (blocking or distorting vessels) but also from the associated inflammatory response they trigger [46]. DCS severity can vary from skin itching and marbled appearance to excruciating pain, convulsions, paralysis, coma and death. Over 60% of symptoms present in the first 3 h post dive, with some presenting as late as 48 h post dive [47, 48], and can be localized (joint pain in a particular articulation) or involve multiple systems. Historically classified as Type I or II for severity, with the second type referring to neurological symptoms, more recently emphasis has been on the progression (or lack thereof) of the disease [47]. In addition to first aid treatment, pure oxygen and intravenous fluids are administered if possible [49, 50]. DCS treatment is to recompress the diver in a recompression chamber to alleviate the symptoms and shrink the bubbles formed, breathing oxygen at high partial pressure to achieve optimal denitrogenation, then bring him back to normobaric conditions. The outcome depends largely on the delay to recompression treatment, in addition to the severity of the hit (for instance cerebral or spinal cord involvement) [51-55]. In a review of 1763 cases, 80% of cases were completely resolved [55].

1.2 Decompression modelling4

Decompression schedules, from either diving tables or most commonly dive computers, are used by divers to manage the risk of developing decompression sickness (DCS). Importantly however, following a decompression schedule, even perfectly, does not guarantee a DCS-free dive. Decompression algorithms dictate the time allowed at each depth before the dive converts from no-decompression dive into decompression dive, as well as the decompression stops needed for a decompression dive (time to spend at various depths on the way up to the surface). These algorithms are calculation principles that follow from a given decompression theory and different ones with wildly different approaches exist.

DCS risk at the moment is relatively well-managed because all the algorithms discussed are fit to real percentages of DCS and therefore "tweaked" to be reasonably safe. However given enough degrees of freedom any algorithm will end up giving the same decompression profile, but this does not mean the "physics/physiology" modelling is correct. Data from the DSL shows that in the recreational setting up to 95% of divers follow their dive computer. It is then essential that dive computer validation is discussed. The current computer validation modalities, although important and most useful as a basic benchmark, still allow for a probability of DCS beyond ideal acceptability in a recreational setting and risk is not well- defined with respect to known physiological factors.

1.2.1 Dive computer validation

This section (1.2.1) summarizes the recommendations of the 2011 Validation of Dive Computers Workshop [56] and in particular the paper “Dive Computers: The Need for Validation and Standards” [57] by Arne Sieber, Milena Stoianova, Ewald Jöbstl, Elaine Azzopardi, Martin D.J. Sayer and Matthias F. Wagner.

4 Adapted from V. Papadopoulou et al., Chapter: Decompression Theory. In: Balestra C, Germonpré P, The Science of Diving: Things your instructor never told you. 2014: 13-36; Lambert Academic Publishing, ISBN: 978-3-659-66233-1. 38

Dive computers have been used extensively in recreational diving for the last 25 years with a low incidence of DCS. It could therefore be argued that their use was “successful” in some respect. However there have been reported cases of DCS, even neurological ones, where recreational divers followed their dive computer on no-decompression dives. The most recent DAN Europe number suggests that around 80% of neurological DCS cases did not violate their diving computer recommendation.

Not many divers realize that at the moment there is no uniform procedure for testing and validating dive computers. They are not even listed under the European Union directive for personal protective equipment (PPE Directive 89/686/EEC). The norm usually applied during the CE certification of dive computers is the EN13319 which only addresses accuracy and precision of the depth sensor and timer. At the moment no dive computer manufacturer provides any details as to the models they use or the implementation of those models and none have ever performed any substantial human validation.

1.2.1.1 Defining “validation”

To develop a validation procedure and guidelines one must first clearly define what the purpose of a dive computer is. In addition to acting like a timer and depth/pressure sensor in real time, divers rely on dive computers for their decompression calculation. That is, to calculate remaining no decompression time at current depth and, in the particular case of either dedicated dive computers or emergency decompression displays, decompression stops during decompression diving. The hidden assumption behind the “trust” each diver assigns to those decompression calculations is that by following the dive computer decompression stops or remaining no decompression time, the diver’s probability of developing decompression sickness (pDCS) is acceptable to him. We therefore all acknowledge in SCUBA diving, quite explicitly when we accept warnings in dive computer manuals or when signing liability release forms to go diving, that pDCS is never zero and that we aim to keep it below a threshold that is personally acceptable to us. However there is no dive computer that will display a pDCS for a specific dive plan in dive plan mode for instance, so the user has to implicitly trust that the recommendation he/she sees on their dive computer display in the form of a no decompression limit or decompression stop has been somehow validated as acceptable for the same type of diving.

The first step in devising validation procedures is therefore to define the “range of applicability” of the dive computer, which will obviously differ tremendously from commercial or military diving in freezing waters at night to recreational no decompression diving in warm waters with good visibility, for example. By clearly defining the range in which a dive computer will be used, precise validation procedures detailing operational needs (display readability in low visibility conditions, temperature sensing and operation, air integration, etc.) and decompression calculations (depth limit with nitrox, or Trimix, etc.) can be outlined.

1.2.1.2 Operational considerations

Operational considerations for the dive computer also form part of the validation process as they need to dictate whether the tool (dive computer) is adequate for use in the predetermined context safely. These include ease of operation of the dive computer, readability of the display in the worst visibility conditions encompassed in the range of expected diving, clarity and unambiguity of the information displayed, obvious failure mode, battery life and ease of displaying and downloading profile data after each dive.

1.2.1.3 Decompression calculations

A method for validating the decompression calculations produced by the dive computer needs to be developed. In this respect, two strategies can be envisaged, depending on whether the dive computer manufacturer clearly states which published and publicly disclosed diving algorithm he is implementing in his dive computer, or not.

In the first case, where the model is published, the validity of the model is not to be proved by the manufacturer which therefore only needs to show that his implementation of said model (in terms of hardware and software design) is faithful to the algorithm published. This would be the most straightforward case since the validity of the model, ie the probability of DCS (pDCS) that the predictions of this model give, are relegated to the developer of the model itself who needs to show how these are acceptable for a specific type of diving range.

In the second case, where the model is unknown, the predictions of the dive computer should be tested against profiles of known probability of DCS (using for instance the US Navy manned dives database) that would be typical for the expected use of the dive computer.

In both cases we come back to the important point of clearly defining the range of applicability of the dive computer to select adequate dives with known pDCS to compare. However it should be noted that since dive computers allow for a “real-time” calculation of the decompression limits, a perfect validation would require an infinite number of profiles to be tested as the combinations are endless. This is obviously impossible and even testing many different profiles is time-consuming and expensive. The other issue is the need to compare to known pDCS dives, which are usually square, table-like dives, not the typical recreational profile. Repetitive, multiday, altitude, gas switches and other common recreational practices are not taken into account using these dives.

Finally, we have been primarily focused on pDCS as the endpoint or comparison point between dive computers’ predictions and known outcomes. The obvious advantage in using pDCS is that we have data to compare to, especially from the US Navy dive computer validation which remains the most comprehensive database in this regard. However using pDCS is not without problems, one of which is whether to include marginal DCS events and those for which diagnosis was uncertain, not to mention that DCS symptoms have a wide range and clustering them altogether might hide different mechanisms at play. In addition, there is the ethical issue with testing pDCS on humans as this is basically inducing DCS in a small fraction of the test-subjects. Scientific ethical approval is difficult to obtain for this - the incidence of DCS among recreational divers is so low that exposing test subjects to a higher risk for validation purposes is considered non-ethical. Using the data already available to estimate pDCS from dives previously made poses a supplementary problem because these come mainly from military test subjects, ie young, male, fit, well-trained, healthy adults which may not be representative of the recreational diving population. This is why the detection of Venous Gas Emboli or VGE to validate diving algorithms has been proposed as an additional endpoint (in terms of reducing decompression stress). Even though it seems intuitively obvious that the more VGE present during the decompression after the dive, the higher the risk for DCS, the presence of VGE does not seem to be a very accurate predictor of the risk of DCS; however, the absence of VGE does positively correlate with a very low to non-existent risk of clinical manifestations of DCS.

1.2.1.4 Proposed lifecycle for dive computers

The following sequence was proposed [57] to be adopted for validating all dive computers:

 Overall scope definition: specify the principal functions of the dive computer, ie parameters to be displayed including display requirements, mechanical design, performance and operational range  Hazard and risk analysis: description of potential risks by fault of diver (exceeding depth limit or no deco time, etc.) or dive computer malfunction (hardware, software, etc.), including an estimate of severity and probability of risk/hazard occurring.  Safety requirements allocation: to limit probability of occurrence and consequences in cases of occurrence of the listed hazards and risks above (clear step with strategies to minimize problems, e.g. clear failure mode display in case of malfunction, etc.)  Design and implementation phase: designing the hardware and software for the dive computer and establishing verification and validation plans  Validation phase: check final product against complete list of requirements, including safety-specific requirements

1.2.2 Brief historical overview of decompression models

1.2.2.1 Dissolved gas phase models: Haldanean-based decompression theory

John Scott Haldane (1860-1936) pioneered modern decompression theory when he suggested that different tissues of the body absorb and release nitrogen at different rates. He proposed four basic principles that agreed with observations [58]:

1. Gas absorption and elimination in a tissue occurs exponentially (assuming that perfusion is the limiting factor) 2. Different tissues absorb and release gas at different rates 3. Decompression is achieved by decreasing ambient pressure 4. Gas tension in a tissue must not exceed approximately twice the ambient pressure

He used five tissue compartments with half-times of 5, 10, 20, 40 and 75 minutes to implement decompression calculations that allowed the development of practical dive tables, first published in 1908. He was also the first to notice that decompression is most dangerous

42 at the surface because of the relative pressure differences, contradicting guidelines of the time which allowed divers to ascend faster as they got closer to the surface.

Haldane's perfusion-limited model was used through to the 1960’s without any fundamental changes (a slower compartment of 120min half-time was added). The 2:1 ratio however was shown to be too conservative for the faster tissues and not conservative enough for the slower ones, when the US Navy started diving deeper than before during WWI and WWII, and the idea that each compartment should have its own surfacing ratio was introduced. As an aside, interestingly the first modification of the 2:1 ratio was to decrease it because it was agreed that only nitrogen played a role in DCS (and not nitrogen and oxygen as originally assumed by Haldane).

Robert D. Workman of the US Navy Experimental Diving Unit (NEDU) revised Haldane's model [59] to take this into account and introduced the term “M-values” to describe how much over-pressurisation each compartment can tolerate at each depth. He also added a further three tissue compartments with longer half-times of 160, 200 and 240 min, and was the first person to present an equation that could be used to calculate the resulting parameters for any depth, a crucial development that would later allow the use of dive computers that have now become a standard piece of equipment in the recreational diving community. The M-value of each compartment is a linear relationship

푴 = 휟푴 × 푫 + 푴ퟎ (1)

where 푀 is the tolerated inert gas pressure in hypothetical tissue compartment, 훥푀 the slope of M-value line, 퐷 the depth pressure (gauge pressure) and 푀0 the intercept at zero depth ie surfacing M-value at sea level. Two interesting practical consequences of this equation are that : 1) the allowable step in decompression is greater at deeper depths and 2) one can surface with some overpressure allowed (M0), leaving the diver with some degree of supersaturation post dive.

Prof. Albert A. Bühlmann (1923-1994) then extended the number of compartments to 16 and investigated the effect of diving at altitude, (i.e. in mountain lakes) [60, 61], after noticing the very high incidence of decompression sickness following dives at altitude using sea-level tables. He developed M-values which, as for Workmann's, expressed a linear relationship between ambiant pressure and the maximum tolerated inert gas pressure in the compartment, 43 the difference being that his were based on absolute pressure, taking into account altitude. The Bühlmann algorithm became the basis for most diving computer programs when it became widely available in 1983-1984 in the book Decompression-Decompression Sickness.

1.2.2.2 Silent bubbles and deep stops: redefining DCS

The traditional Haldane approach to decompression and up to its Bühlmann continuation, described in the above section, was challenged in the late 1960s when Merrill Spencer first showed, using Doppler ultrasound, that bubbles were present even in DCS-asymptomatic divers [15, 19]. This challenged the Haldane-Bühlmann models as they were based on the assumption that bubble formation only happens when an M-value is exceeded. Because these divers all dove well within the limits of the decompression models and showed no signs of DCI, those bubbles are commonly referred to as “silent” or “asymptomatic” bubbles. Their existence however led to the development of new decompression models because of their association with repetitive diving.

Typically these silent bubbles would not interfere with the normal functions of the body and slowly diffuse back into the blood in the alveoli of the lungs so the gas is expired out. However they do need to be taken into account for repetitive diving because the subsequent dive starts with some bubbles already present and therefore requires the models to be adjusted. Silent bubbles were also linked to fatigue in the surfacing diver [62].

In addition to this practical consideration however, the existence of silent bubbles also challenged the Haldane-Bühlmann models on a much more theoretical ground, showing that the belief that bubble formation equates to DCS is wrong (liquids such as blood will hold gas in a state of supersaturation provided the supersaturation does not exceed a critical limit). A new definition of how many/how big silent bubbles need to become to cause DCS was needed.

In Haldane-Bülhmann models the decompression algorithm basically calculated how shallow a diver is allowed to ascend to, so as not to exceed his M-values for the different compartments. The diver then ascends at the appropriate rate to the allowed depth and waits the required amount of time before ascending again to the next allowed depth, etc, until the surface can be reached without exceeding any M-value. Theoretically this not only ensures

44 that the diver does not get DCS but also minimises the time spent decompressing (which is the aim of every decompression model). In other words Haldane-Bühlmann models are characterised by relatively fast initial ascents.

The original Haldane model had already been modified several times and in practice these modifications included the introduction of deeper initial stops for decompression [63]. These were shown to significantly reduce the amount of silent bubbles formed. Richard Pyle introduced those deep stops (also known as Pyle stops) from his personal experience. Pyle would collect deepwater fish for the University of Hawaii and, in order to keep them alive, he was required to stop at around half the depth to depressurize the swim bladders of the fish. (Figure 4) He considered how tired he felt after his dives and concluded that the deep stop performed helped him. In practice his method consisted of making an additional stop between his maximum depth and his first required decompression stop (at half the distance between them if that distance was less than 9m or repeat the process if greater than 9m). The benefit of deep stops is debated today, with evidence that this merely shifts the decompression stress further into the slower tissues, thus potentially increasing the DCS risk for some dive profiles (Figure 5).

Figure 4 : Richard Pyle puncturing the bladder of a fish while ascending from the depth.

Pyle's empirical approach then evolved into the Gradient Factors (GF) theory which aims to reduce silent bubbles within the Bühlmann model. The basic principle of this is to increase safety by staying further away from the M-values. Gradient factors are expressed as percentages between saturation and supersaturation, ie a 0% GF is equivalent to the saturation line, whereas a 100% GF is the M-value line. In practice gradient factors are used in pairs with a lower GF (Lo) and upper limit GF(Hi) usually set to 30 and 80 % respectively.

Figure 5 : Supersaturation of fast and slow compartments for a shallow and deep stop profile. A shows the two 30min air decompression profiles tested; B the supersaturation of a fast (halftime=10min) compartment for these two profiles; and C the supersaturation of a slow (halftime 160min) compartment for these two profiles.

1.2.2.3 Tracking free-phase gas (bubbles): dual phase models

The traditional decompression models are sometimes called “dissolved gas models”, as they are based on the assumption that inert gases are held in the tissues in solution form until they exceed their M-values (supersaturation) and form bubbles which cause decompression sickness. As we have now seen, this cannot always be the case since the presence of silent bubbles challenges this view. Dual Phase Bubble models were developed as an alternative to take into account the Doppler ultrasound findings.

In the early 1970s Brian Hills, an Australian PhD student, developed a “kinetic and thermodynamic approach to decompression sickness” [64], tracking not only the dissolved gases in tissues, but also the gas in the form of bubbles (thus the name Dual Phase). His ideas were initially criticised and it was only much later that others followed his research, in particular Val Hemplemann and Tom Hennessy, who suggested the introduction of a critical limit to the total volume of free gas phase gas as a definition for DCS, and the Tiny Bubble Group (David Yount's research group in Hawai) who investigated the factors influencing bubble formation.

Dual Phase Bubble models (an example of which is the one developed by Hills and Le Messurier advocating a very low level of supersaturation) track two parameters, the free gas phase (bubbles) and the dissolved gas phase, and in a similar manner to the traditional models they consider a number of compartments with different half-times that determine the rate of uptake and release in and out of each compartment. Decompression is then achieved by controlling the total volume of free gas while ascending. In practice this is achieved by calculating the supersaturation gradient that will ensure that only bubbles above the critical radius are allowed to grow, thus limiting their number and tracking their total volume. The main assumption of this model is the statistical distribution of bubble sizes.

The Varying Permeability Model (VPM) [65, 66] is a dual phase model with 32 compartments, 16 of which track Nitrogen loading, and 16 of which track Helium loading. As a dual phase model it does not monitor M-values and instead limits the ascent through bubble considerations. The assumption of VPM is that micronuclei (gas seeds which permit the formation of bubbles) are present in the body at all times, even before dives and for non- divers. These are small enough that they are in a solution form but do not get completely crushed under pressure (note that at a depth of 80m this view has now been challenged).

Originally the VPM based its decompression algorithm on bubble number so that no more than a certain number of bubbles were formed. However this was shown to be way too conservative for non-saturation diving and led to the development of a Critical Volume algorithm that instead limits the total volume of free gas during decompression:

푻풐풕풂풍 품풂풔 풗풐풍풖풎풆 ∝ 푮풓풂풅풊풆풏풕 × 푻풊풎풆 × 푵풖풎풃풆풓 풐풇 풃풖풃풃풍풆풔 (2)

The basic assumption is that DCS only occurs above a certain threshold of free gas volume. This model was further developed into VPM-A between 1999 and 2001 by Dr David Yount, Erik Baker and Eric Maiden (Hawaii University) to incorporate repetitive diving, multiple inert gases and gas switches (changing gases underwater), which quickly became very popular within the technical diving community as it was available for free on the internet. A fundamental flaw in the model was corrected in 2003 by Erik Baker (this new version is referred to as VPM-B) who noticed that VPM-A only calculated a supersaturation gradient at the start of the ascent and did not update it for every depth. In 2005 the model was made more conservative by Ross Hemingway who also developed a popular computer program for calculating decompression stops (the Z-Planner and, later, the V-Planner), reducing the allowed gradients in shallow depths for difficult dives as an extra-safety margin, if selected by the user (VPM-B/E).

The Reduced Gradient Bubble Model or RGBM developed by Bruce Wienke claims to deal with some of the known risk factors for DCS such as water temperature, flying after diving and altitude diving, but also repetitive diving and mixed gas diving. The RGBM follows the VPM model in its logic and is, as such, also a dual phase model. A set of bubble factors are used to take into account the conditions of the dive (one for repetitive diving over hours, one for repetitive diving over days, one for reverse profile diving, etc). Each bubble factor is then applied to the base level gradient to adjust the decompression profile.

1.2.2.4 Alternatives and challenges to modelling assumptions

In addition to the models described above, there exist other models, of which we will only present here the basic assumptions and differences to the RGBM for the most popular.

RNPL - diffusion limited

All the models discussed so far are perfusion-limited models in that the assumption is that it is perfusion to the tissues that controls the uptake of inert gas. The model developed by the Royal Naval Physiological Laboratory (RNPL), based on the Hempleman tissue slab model, challenges this view by considering the diffusion rates of inert gas from the capillaries to the tissues. The most current version of this model is the BSAC-88 which, incidentally, is the model taught to British recreational divers.

US Navy “Thalmann algorithm” model

Captain Ed Thalmann was the original developer of this method commissioned by NEDU to be incorporated in a dive computer algorithm. The main difference between this method and the RGBM is that, although the on-gassing of tissues is considered to be exponential, the off- gassing is linear to limit the silent bubbles detected with Doppler ultrasound. This model was incorporated into dive computers used by Navy SEAL teams in 2001.

Probabilistic

There have been several attempts to incorporate a probabilistic approach into current decompression models [45]. The current probabilistic approach is based on the binomial distribution, considering a no/yes approach for DCS occurrence in the databases used for calibration, and was used to update and extend the Thalmann algorithm for the US Navy. In practice, it is then possible to calculate a probability of DCS occurrence for a particular profile from the calibration database used. Probabilistic models also rely heavily on decompression models and their parameters (as deterministic models do). One should keep in mind that these do not account for different physiologies and fitness levels, as the recreational diving population is much more diverse than the typical 20-something fit navy divers that tested the models for the most part.

DCIEM

This model was developed by the Defence and Civil Institute of Environmental Medicine (DCIEM), Canada [67-69]. A key difference between this model and the RGBM is that the compartments considered are not seen as independent from each other but also diffuse to and from one another. The current version of this model was published in 1992. It is viewed as one of the safest tables, with over 5000 dives and VGE grades having contributed to its 49 testing and development. It is interesting that in the development of these other decompression models one of the recurrent tweaks that had to be performed was, in practice, equivalent to incorporating deeper stops. Human trials are, however, yet to confirm if current deep stop practices are an improvement or, in some circumstances, an additional risk. Even so, for the majority of recreational/technical diving, the tables produced using these different models are very similar despite their theoretical differences. This is because the parameters of all models have been greatly adjusted to fit observational data.

Figure 6 : A horizontal diver during decompression

1.2.3 Conclusion

In recreational diving, DCS risk seems to be relatively well controlled, mainly because decompression algorithms have been fitted to real data of DCS dives. Nevertheless, both the physics and physiological changes associated with SCUBA diving are not fully understood with respect to the potential of developing DCS. In particular “technical diving” and other dives pushing the limits of the algorithms continue to have a higher risk of DCS. New research is needed into a personalised decompression algorithm that would take into account 50 physiological factors and inter-personal differences of decompression stress. In this respect having appropriate validation tools both for the implementation of decompression models in dive computers, but also for validating the actual algorithms used, would be greatly beneficial.

Chapter 2 – Literature review and PhD aims

2.1 A critical review of bubble formation in hyperbaric decompression5

2.1.1 Aim and scope

The study of bubble formation and growth in hyperbaric physiology and the factors which influence them is of prime importance for understanding the pathophysiology of DCS and improve its prevention and treatment. Echographic recording and imaging of bubbles has shown the number of bubbles post dive to be an indicator of decompression stress. It is as such a good way of improving DCS prevention by devising decompression schedules which control the number and size of bubbles formed, instead of relying solely on the outcome DCS/no DCS to quantify success of the decompression schedule [11, 15, 19].

The study of bubble formation can improve preventive measures against DCS risk in two ways. Firstly by improving the decompression algorithms which rely on bubble modelling and secondly through predive conditioning [70, 71] that would target bubble growth itself. Together with DCS studies, it is also relevant in physiology, in particular to understand the processes of adaptation to extreme environmental stress, but also for hyperbaric oxygen (HBO) treatment and as a study of tolerated embolism to the circulation which can have medical applications (ultrasound microbubble contrast agents or surgical and mechanical ventilation embolism risks).

The review in this Section 2.1 aims to cover the literature on physiological bubble formation during decompression, primarily in the context of scuba diving. This complex research area links a variety of disciplines with drastically different methodologies ranging from mathematical modelling to physiological studies. As such, a comprehensive study of where

5 Adapted from V. Papadopoulou et al. Adv Colloid Interface Sci.,2013;191–192(0):22-30. 53 these agree and disagree would be useful in summarizing the limits between theory and observations from experiments in vitro, ex vivo and in vivo. This critical review provides an up to date list of references for this field of study. The current consensus and disagreements in the field are pointed out and the successes and limitations of the studies included are discussed. Where appropriate, suggestions for further studies to be undertaken are included. The relevant physics background of bubble formation (nucleation) is also included.

2.1.2 Background

2.1.2.1 Fundamental physics

Supersaturation can be viewed as a tissue's tendency to produce bubbles and as such depends on the difference between the gas tension in the tissue and the ambient pressure. Supersaturation normally results from a saturated solution being subjected to a thermodynamic change which increases its concentration further (thus bringing it beyond saturation) such as an increase in temperature, decrease in volume or decrease in ambient pressure. In the scuba diving context, pressure is the main variable of interest. The diver's tissues become saturated in inert gas at depth. For a diver breathing pressurized air (about 21% oxygen and 79% nitrogen), this inert gas is nitrogen since most of the oxygen is being used by the body through metabolism [26]. During ascent, the tension of the dissolved gas in the tissues is greater than its partial pressure in the lungs.

Nucleation is the localized formation of a new thermodynamic phase out of a solid, liquid or gas phase. The case of interest here is the formation of bubbles (gas phase) from tissues (assumed “liquid” phase). There are two types of nucleation: homogeneous and heterogeneous. The most common nucleation process is heterogeneous nucleation where nucleation happens at specific sites between two phases or around microscopic impurities. Homogeneous nucleation happens where nucleation does not have preferential sites for the bubbles to grow from. The random fluctuations of the molecules in the liquid are statistically likely to form microscopic regions where molecules are more closely packed together and voids where the bubbles grow from. As this process is random such sites are created throughout the liquid and thus no preferential sites exist [72]. This process is actually not very common since bubbles are normally observed to nucleate from preferential sites and homogeneous nucleation usually involves supercooling or superheating. It seems therefore 54 unlikely that homogeneous nucleation is at the origin of the venous gas emboli observed in these conditions.

Tribonucleation [73] is the formation of new gas bubbles in a solution where two adhesive surfaces are rapidly separated from one another due to the resulting negative pressure that ensues momentarily [73-75].

In the context of the current review, cavitation is defined as the process of bubble formation from a nucleus. This can occur either when the pressure in the fluid drops below the saturation vapour pressure (boiling cavitation), or due to the desorption of dissolved gases (degassing cavitation) which can happen at a higher pressure than the saturation vapor pressure [72].

For a bubble to form “spontaneously” in a solution with dissolved gas a supersaturation greater than 10.0 MPa is required which is reached at about 900 m sea water (msw). However bubbles have been observed by ultrasound imaging in divers after dives as shallow as 3.6 m (equivalent to about 40 kPa of pressure) [76]. De novo bubble formation by homogeneous nucleation does not seem possible from the pressure excursions of human decompression [25, 77].

Heterogeneous nucleation however can account for bubble formation in relatively low supersaturation levels such as the ones observed in the scuba diving context. This is at the origin of the concept of “micronuclei” which are hypothesized to act as gas nuclei or “seeds” for the bubbles to grow from [25, 77]. Their current definition is small gas filled bubbles whose size does not exceed 10 μm [78]. For bubbles to grow from micronuclei, the dissolved gas in the supersaturated tissues entering the micronuclei needs to overcome its surface tension. The smaller the bubble is, the stronger its surface tensile strength. For a given pressure gradient there will therefore be a critical size radius above which the bubbles will be excited to growth.

2.1.2.2 Early Studies

The postulate for micronuclei has actually been around for a long time, with little direct experimental evidence to support it in vivo. Leonard Hill described how bubbles need weak “points” to grow from a fluid as early as 1912 [79]. In the 1960s and 1970s Brian Hills 55 developed an approach to prevent decompression sickness (which he described as “thermodynamic and kinetic”) focused on tracking both dissolved and free phase gas [64]. Hempleman and Hennessy suggested that a certain degree of embolism is tolerated by the body and suggested a new definition of DCS as a volume threshold of total gas bubbles circulating [80].

In 1954, Herzfeld and Fox [81] discussed the necessity of nuclei to induce bubble formation via ultrasonic cavitation, arguing that pressures in the order of hundreds of atmospheres (tens of MPa, i.e. below 900 msw) would be needed otherwise. These bubble seeds in turn need to be stabilized since their surface tension alone would otherwise dissolve them. To account for this they suggested bubbles with organic skins [81]. These stabilized bubbles may then act as cavitation nuclei. This is also supported from Harvey's work [77] whose experiments showed that undissolved gases play a key role in such cavitation. No cavitation could be ultrasonically induced for some time after subjecting water to 1000 atm for 15 min and rapidly decompressing it. Harvey explained this observation saying that the high pressures forced the gas into complete solution and today this would be called a “denucleation procedure”.

In addition to Harvey who showed that fissures in solids could support growth of free phase gas, stabilization through geometrical considerations, looking at contact angles of the growing bubble from a non-flat surface, was also considered by Greenspan and Tschiegg [82] who showed that the acoustic cavitation threshold in water could be raised by passing samples through membrane filters.

In 1968, Campbell evaluated quantitatively the theory that tribonucleation in liquid solutions happens in areas of reduced pressures as two attractive surfaces are rapidly separated, and discussed the role of the solid surface composition in such homogeneous nucleation [73]. He found that to induce bubble growth to a macroscopic size, gas in solution had to be present, similarly to the cavitation conditions described by Harvey and Herzfeld. The contact angles that would yield bubble formation were also discussed as surface crevices of the separating surfaces were considered geometrically.

The stability of gas bubbles in liquid solutions was analytically treated by Epstein and Plesset in the 80s [83], first using diffusion theory and then more systematically in full thermodynamic considerations, resulting in the well-known Raleigh–Plesset equation. A

56 series of experiments that supported the gas nuclei concept were performed in the late 70s and early 80s by the Tiny Bubble Group (Yount et al.) in which they pressurized and then decompressed transparent gelatin to study bubble formation [65, 84, 85]. The motivation for this was the simple observation that DCS can present in almost any part of the body. They hypothesized that this is due to the properties of water relating to cavitation. Gelatin was chosen as an aqueous medium because it is conveniently transparent and the bubbles produced in it stationary which makes it easy to count and size them optically. Their series of experiments in 1976 concluded with the hypothesis that bubbles form out of pre-existing nuclei. Bubble formation in humans and in supersaturated distilled water was shown even for pressures below 1 atm, whereas the calculated tensile strength of water should exceed 1000 atm [84, 86]. This cannot be explained by “solid impurities with smooth surfaces” [87]. Moreover the cavitation threshold was shown to increase significantly after degassing procedures, a specific test for gas nuclei. A similar technique was tried on the gelatin samples where static pressure was applied and the cavitation threshold for gelatin was shown to increase. Yount et al. concluded from these observations that gas nuclei were probably at the origin of bubble formation in scuba divers and they went as far as suggesting denucleation procedures from pressure excursions or via drugs. They hypothesized that the acclimatization observed in many caisson workers [88] might be due to micronuclei population depletion. They also urged for new decompression algorithms that would incorporate bubble dynamics directly (instead of compartment ratio considerations only) to be developed in light of these findings. They suggested asymptomatic bubble outcome of a dive measured ultrasonically as a way to measure the success of the decompression schedule since all dives result in some degree of bubbling [17, 89].

Some of the early experimental “evidence” for the micronuclei concept in vivo came from Evans who in 1969 showed that decompressing shrimp after compressing them to 400 atm resulted in significantly less bubbles [90], presumably from “crushing” the micronuclei population pre decompression. Other animal experiments included the decompression of rats breathing air after a very short pressure excursion to 3 MPa (305 msw) before leaving them at 0.7 MPa (73 msw) for 2 h. This experiment showed significantly less DCS occurrence for the rats which went to 3 MPa with respect to those which didn't [91]. Both studies seem consistent with the idea of shrinking gas bubble precursors before decompression to account for less bubble formation, thus supporting the micronuclei theoretical basis. They also show potential for preventing DCS by developing procedures to target micronuclei pre dive. 57

2.1.2.3 The micronuclei stability problem

The concept of micronuclei is not without its problems. The most important one is accounting for the long term existence of these gas filled microbubbles since from their tiny size it would be expected that they spontaneously shrink in the absence of other stabilizing forces, for instance surface active coatings. A stabilization process to account for non-negligible half- lives therefore needs to be discussed [76, 77]. To understand the problem let us consider the three forces acting on a single bubble. Those will be the gas content pressure pushing outwards, the ambient pressure pushing inwards and the Laplace surface tension of the bubble. For a bubble to be stable they need to equilibrate. The surface tension force is inversely proportional to the radius so becomes dominant for very small bubbles that should have a tendency to dissolve below a critical radius at ambient atmospheric pressure. Micronuclei, of the order of a couple of micrometres, need to be stable for ambient atmospheric pressure but without invoking an additional stabilization mechanism they would dissolve immediately. Mechanical stability is a necessary but not sufficient condition for bubble stability. For a microbubble to be stable, it also needs be thermodynamically stable or in other words in chemical equilibrium with its surroundings. Since micronuclei are hypothesized to exist independently of diving, they should be stable at atmospheric pressure. However in the absence of a stabilizing mechanism, their surface tension would shrink them to dissolution [92].

One way of overcoming this issue is to invoke surfactants which are amphilic organic compounds that lower the surface tension of bubbles [65, 93, 94]. The Tiny Bubble group first looked at surfactant molecules as stabilizers [65, 87]. The permeability of the membrane of the bubble which depends on the diffusivity across the bubble then also dictates the rate of bubble growth. However, a surfactant that would lower the surface tension term of the Young Laplace equation has not yet been identified in vivo [25]. In addition, in vitro studies showed the opposite effect with known surfactants, as an increase in surfactant induced a decrease in bubble formation after decompression [95]. Another stabilization mechanism comes from geometrical considerations to find contact angles that would permit bubble growth. For instance hydrophobic crevices have been suggested.

Yount et al. proposed a stability mechanism [65] for gas nuclei. This is needed to explain why stable nuclei seem to exist which is a priori counter intuitive since large gas phases with a radius above 1 μm should rise to the surface of a standing liquid and smaller ones should 58 diffuse into the surrounding liquid due to surface tension effects. Herzfeld and Fox's idea of an organic impermeable skin to stabilize bubble nuclei was abandoned after Strasberg showed that a cyclic change in pressure did not leave the nuclei unaffected [96]. In addition the counter diffusion that seems to happen where multiple gases are involved [97] also goes against impermeable skins. To overcome the problem encountered with impermeable skins, Yount et al.'s hypothesis was a stabilization mechanism based on surface active skins of varying gas permeability. The idea is that the surface of the bubbles must be initially permeable for the gas to diffuse inwards, then should progressively become effectively impermeable above a threshold static pressure applied rapidly. These properties were found in practice to be similar to having surfactant skins so this model became the surfactant stabilization theory.

2.1.3 Recent Studies and Discussion

2.1.3.1 Bubble formation mechanisms

In 2008 Goldman revised and lowered the pressure threshold needed for homogeneous nucleation. Applying a similar approach to Abraham's thermodynamic study of liquid droplets surrounded by vapour phase [98], Goldman derived Gibbs free energy expressions for gas bubble formation from supersaturation [99]. The nucleation energy threshold was shown lower than previously thought resulting in the theoretical possibility for homogeneous nucleation to occur for human decompression situations (less than 5 atm, roughly equivalent to 40 msw) in very particular tissues with very low surface tensions. The Gibbs free energy expressions obtained are based on the assumption that the system under consideration is “closed”. Physiological tissues however are perfused by blood and exchange dissolved gas. Therefore this study can only be applied to physiological tissues under the assumption that this exchange with the circulatory system is slow enough to be ignored during the nucleation process. In other words there is a “separation of time scales” between perfusion and nucleation processes. As pointed out by Goldman however, even if homogeneous nucleation did actually occur physiologically it would still only account for a very small percentage of the venous gas emboli observed ultrasonically, especially since bubbles are detected with pressure exposures far smaller than 5 atm [11]. Therefore the conclusion of this study in the scuba diving context, assuming that the approximations hold, still maintains that

59 heterogeneous nucleation, tribonucleation and bubble growth from stabilized pre-existing micronuclei are definitely more important processes than homogeneous nucleation.

A potential candidate for micronuclei was discovered, as atomic force microscopy has shown that gas nanobubbles of 5–30 nm form spontaneously on smooth flat hydrophobic surfaces submerged in water [100-104]. As suggested by Arieli in 2011 [105], hydrophobic surfaces in the body, for example in large blood vessels and fat, might therefore be where micronuclei are formed, without necessarily having crevices. To support this hypothesis they looked at the formation of bubbles on hydrophobic and hydrophilic smooth silicon wafers in degassed water compressed to 90 m for 15 h then decompressed. The results showed bubbles only on the hydrophobic surface. The possibility for these nanobubbles to act as nucleation sites was discussed in other studies, showing them to be so stable (“superstable”) as to exclude the possibility they would act as gas nuclei for bubbles to grow from them. Their stability was demonstrated not only for ambient pressure but also for a reduction in ambient pressure down to 6 MPa [106]. These nanobubbles have therefore been shown to be stable for hours, despite the expectation that they would dissolve in much less than a second due to their large Laplace pressures. This superstability has not been explained theoretically as yet. However it should be noted that this stability during huge pressure fluctuation does not exclude growth by gas diffusion from a supersaturated tissue and in Arieli's et al. study, probably mainly due to the very high supersaturation of the tissues, the nanobubbles do appear to act as micronuclei. The critical radius of curvature of these bubbles is 100 nm above that for which they can evolve as bubbles [103]. This seems to be in agreement with the evolution on hydrophobic crevices. Higher percentage of adipose tissue is a known predisposition to DCS. This was traditionally explained through the fact that, being more aqueous, it was a medium in which nitrogen dissolved better. This theory might offer an additional if not alternative explanation: that hydrophobicity of adipose tissue makes it a preferential site for bubble growth from a larger micronuclei population from nanobubbles.

2.1.3.2 Stability

The stability issue of micronuclei was investigated further and an alternative solution, proposed by Goldman in 2010, avoids the issue of surfactants not having been identified in

60 vivo [107]. Mathematically the Young–Laplace equation for a spherical stable gas bubble was generalized [108] to include effects of its surface tension and elastic forces from its surroundings, assumed to be a soft isotropic material. The resulting Generalized Young Laplace equation (GYL) is exact in the regime where these spherical bubbles are large enough (above 1 μm) to ignore microscopic behaviour of the surface tension and the interface between the bubble and its surroundings. The Gibbs free energy of deformation of the elastic surrounding is also derived. Treating the tissues as soft deformable materials in a supersaturated state, the Gibbs free energy for the system and per bubble was derived using the GYL equation derived in 2009 [108]. For a material with a non-negligible shear modulus, this demonstrated that free energy wells which would stabilize small gas bubbles exist. A new model for tissues as “isotropic elastic materials that have a surface tension and resist both compression and shear forces” is thus proposed, which would account for the micronuclei hypothesis while solving their stabilization problem. Tissue elasticity is therefore considered here as a potential explanation for micronuclei stability that accounts for both mechanical and thermodynamic stabilities. The derivation shows that bubbles below a certain radius and above a certain radius will be mechanically stable, whereas any radius size in between will be unstable, because of the opposite forces acting on the bubble pressure due to the surface tension and shear resistance. These thresholds depend on the relative magnitude of the shear modulus of the elastic material under consideration, but they correspond in size roughly to radii below 0.8 μm and then above 6 μm. This interesting property could explain why very small bubbles (bubble nuclei) need the dissolved inert gases to trigger their growth above the radius for the second mechanically stable region. Another study combining this mechanical stability to chemical stability considerations found that elastic materials with non- negligible shear modulus can indeed yield stable micronuclei for the bubbles to grow from [107]. The main potential for criticism in this study comes from the relatively small stabilizing potential wells which can account for bubble stability over finite periods of time (short times, but actual timescale not given). However assuming that tribonucleation happens in the body from muscle movement for instance, which has been suggested as a mechanism to explain the higher DCS occurrence after exercise [25], this could account for a non- negligible population of micronuclei at any one time as new micronuclei are formed by tribonucleation.

An alternative stabilization mechanism, from geometrical considerations, was considered. A bubble growth model from hydrophobic conical crevices inside vessels was tested with 61 realistic tissue parameters by Chappell and Payne [109]. They looked at their behaviour under compression [110], showing that the geometry would resist the compression via slight deformation and change of radius of curvature. They then investigated how bubbles could grow from these crevices under decompression [111]. The model was developed to account for a single inert gas and gas transfer happened through the walls of the crevice. Incorporating metabolic gases (in this case oxygen) was shown to have a measurable impact in making the surface tension less significant in the nucleation rate. This was explained through the high diffusivity of metabolic gases. Since hypobaric decompression yields bubbles with a greater percentage of metabolic gases as observed in astronauts [112, 113], it would be interesting to check this theory by applying the model to a hypobaric decompression scenario. The cavity geometry was also looked at and four different geometries analysed [114], while neglecting gas transfer. The nucleation behaviour was found to depend mainly on the size of the mouth of the cavity after initial growth where the bubble reaches the opening. At this point the flow conditions also play an important part, as one might expect.

2.1.3.3 Single and multiple bubble behaviour

Single bubble growth on a solid surface was studied by generating a single bubble on a submerged heater. As heat is applied the liquid becomes supersaturated locally. This procedure is somewhat different to the traditional way to study these phenomena by degassing supersaturated liquid through decompression, where a single bubble alone cannot be produced. Thermal degassing which involves mass transfer but also heat transfer is thus achieved (as opposed to decompression degassing which only involves mass transfer in theory). To study the bubble generation and growth separately from the gravitational effects on them, the experiments where performed in microgravity conditions (ESA parabolic fights) [115, 116]. The experimental observations were compared to a theoretical model [117] derived considering spherical bubbles in a uniformly cooled liquid that were heated from the inside. The initial stage of growth was shown to agree with a parabolic diffusion law, after which a linear growth model was more appropriate. The lateral motion of the bubbles along the heater as they are first generated was also looked at and discussed with respect to the surface of the heaters used [118]. Multiple bubble growth and detachment showed competition for the dissolved gas available in the supersaturated solution amongst bubbles

62 growing closely together [119]. The final size of bubbles was shown to be smaller than that of a single bubble, and a critical temperature could be found above which any increase in temperature did not result in faster bubble growth.

Karapantsios et al. have argued for the necessity to study the characteristics of bubbly flow (multiple bubbles flowing with the liquid) in addition to single bubble generation, since it is this abundance of bubbles which is at the origin of DCS above some threshold. An impedance spectroscopy technique, In Vitro Embolic Detector (IVED), was developed to detect bubbles in the blood stream by measuring the gas fraction. The in vitro phase of this project showed very good resolution as well as sensitivity to variations in gas fraction and bubble size in bubbly flows [120], and the in vivo phase is in progress. The results were validated through acoustic spectroscopy and electrical impedance tomography measurements. An in vitro experiment to simulate a realistic bubbly flow in the human vena cava was devised to investigate the effect of surfactant and/or electrolyte concentrations on the bubble size distribution (measured both by the IVED and electrical impedance tomography for validation purposes) to continue the improvement of these techniques but sized optically in this study [121]. The study found no correlation between the bubble size and the radial position in the tube or viscosity of the liquid. The size distribution was however found dependent on the flow rate and lower for higher surfactant and electrolyte concentrations and when both were added together this effect was amplified. An assumption throughout the paper is that the addition of surfactants will not affect the radial distribution of different bubble sizes in the tube, and all measurements for sizing were done near the surface of the tube. Another limitation of the study is the high bubble count needed for sizing.

A mathematical study to look at the interaction between blood born bubbles and tissue bubbles was conducted, assuming that bubbles can form in tissues and in the wall of vessels [122]. Once again competition for dissolved gas was pointed out. It was also shown that the number of tissue bubbles will influence the number of blood bubbles, whereas the opposite effect is very unlikely. The main variable of interest is obviously perfusion of tissues.

The phenomenon of competition for dissolved gas among growing bubbles was further investigated through numerical simulations [123]. A clamping phenomenon was demonstrated above a certain density of bubbles per unit tissue, after which the washout rate was considerably diminished, going from exponential to linear. This finding seems realistic since a number of decompression algorithms, the so-called exponential linear kinetics 63 models, use a linear washout rate with very good correlation to real dive data [124]. The main limitation of this study is the lack of information on in vivo bubble density in tissues which makes it very difficult to extrapolate the findings.

As an aside, the possibility for a single bubble to act as a “gas plug” was found possible [125], after calculating what size a bubble would need to be to block some of the capillaries in the body. For almost any driving pressure difference affecting bubble growth this was shown to be a possibility. The compliance of the vessels was not taken into account in the derivation of this model, however, and it would be expected that their contribution to this problem will be significant. It would be interesting to look at incorporating this, as a systematic analysis of the gas plug possibility resulting from a hyperbaric exposure was never investigated before Chappell and Payne's study.

2.1.3.4 Role in decompression modelling

Although the review in this Section 2.1 has been primarily focused on bubble nucleation in the physiological context of human hyperbaric decompression, it is important to note that bubble growth is obviously also very relevant. To develop a decompression model based on physical parameters, both nucleation and growth have to be described, giving the rate of bubble “appearance” and that of “growth” respectively. Combined, they would allow for the precise calculation of bubble size distributions with respect to dive time, which could then be checked. In vitro physics experiments can thus be used to determine which parameters influence and dominate bubble number and size (nucleation and growth phases), as physiological studies can only observe their combined effects. In particular, exploiting the controlled set up from in vitro experiments can allow to study isolated phenomena, decoupling heat, mass transfer, gravitational and/or bubble competition effects as seen in the previous section.

Explicit bubble dynamics have been incorporated in the modelling of decompression to produce safer decompression procedures. They use Venous Gas Emboli (VGE) as a way of evaluating decompression models instead of only relying on the incidence of DCS. The Copernicus model [126, 127] includes bubble considerations and assumes that bubbles grow from pre-existing gas nuclei as Yount suggested in 1979, then relies on an approximate

64 stabilization function from Yount 1979 and Chappell 2006. It also assumes that gas nuclei are attached to the endothelial layer [128].

A modification of the Srinivasan et al. model (1999) was presented in 2010 [129], where the concentration distributions around a tissue bubble at ambient pressure after decompression was solved analytically to find its growth rate. This was shown proportional to the ascent rate, tissue diffusivity, initial concentration differences and void fraction (dominant factor) and inversely proportional to the surface tension. The calculations were solved analytically for unsteady flow in tissue based on the three region model [130], showing how the concentration gradient decreases as the bubble grows.

A biophysical model specifically aimed at articular bends was developed recently [131]. The joint was separated into two compartments exchanging inert gas via blood perfusion and with one another via diffusion. The diffusion interface, along with relatively large diffusion coefficients, could account for the late onset of symptoms often observed. A critical volume of free phase gas was used as a definition for DCS. A clamping phenomenon was observed soon after the decompression onset. The model fitted the data well.

The need for a realistic biophysical model for bubble growth during hyperbaric decompression has been made clear with studies showing that the extrapolation region for dissolved gas models was not particularly good: if the models do not mimic biophysiological processes enough they cannot be extrapolated to situations different to their calibration dataset. However such a satisfactory model has yet to be developed [132]. In that respect using VGE counts instead of DCS/no DCS outcomes only is particularly important. This was made even clearer via a study on mild or “marginal” DCS events [133]. The difficulty arising from marginal DCS events (for instance skin rash) to calibrate probabilistic decompression algorithms to dive data has been discussed recently [133]. They were traditionally assigned fractional weights, resulting in more conservative models. A statistical analysis was performed to see whether these could be described as random occurrences, in which case they should not be included in model parameter fitting. Interestingly the study concluded that these should not be included in model fitting, since model calibration without them yielded the same correlation coefficient and similar extrapolation regions to real dive data. Analytically the calculated weighting that should be applied to these events was found to be 0. This highlights the difficulty of looking at DCS outcome as the sole indication for the calibration and validity of a model. Using VGE scores is much more powerful. The role of 65 decompression models is no longer to limit DCS occurrences, it is to limit VGE scores post dive.

The current practice in VGE monitoring post dive relies on trained observers, usually clinicians, attributing a severity grade to the Doppler ultrasound video they have from the heart (or audio recording). Different scales exist (Spencer or Kisman-Masurel (KM) for sound recordings, and Ikeda, Eftedal–Brubakk (EB) for video) with 4 or 5 severity grades [11], but all rely on the frequency and amplitude of the signal, in other words number of observed bubbles per cardiac cycle as well as the relative intensity with respect to the cardiac sound in the case of audio recordings. However current grading methods have been shown to be inconsistent [134] as they are user dependent and the monitoring times post dive are not consistent between studies. In this respect having an objective, quantitative VGE scoring system, ideally a bubble counter and sizer per unit volume and time would be very useful, and indeed efforts are being made in this general direction [135-142].

Modelling considerations also include finding ways to explain physiologically the influence of known risk factors on bubble counts observed, such as exercise [143] and immersion [144]. An interesting modelling attempt in this direction to predict the median peak bubble grade post-dive evaluated by Doppler ultrasound in controlled physiological conditions [145] combines a dissolved gas phase model [146] with a bubble dynamics model for perfused tissues [123].

The evasive influence mechanism of exercise on bubble counts measured post dive was also investigated theoretically [143]. The bubbles were assumed to follow a Poisson distribution of formation with respect to time and their growth only dependent on pressure differences. Exercise was then factored in through an elevated consumption of oxygen and enhanced perfusion of tissues. This was shown to result in longer lifetimes for tissue bubbles and less bubble growth overall. However at the same time tissue motion could increase bubble formation through motion induced cavitation. To accurately predict outcome the relative rates of these processes would need to be calculated, which is in practice very difficult as there is no way of quantitatively characterizing these separately. Exercise timing and intensity, as well as the nature of exercise, which have all been shown to give particularly different results, were not considered systematically in this study.

A recent study showed that VGE bubble counts were significantly (p < 0.0001) higher for in water diving compared to the same dive profile in a dry chamber [144]. This is particularly important with respect to testing and parameterizing models which increasingly rely on venous gas bubble counts. In particular, deciding whether to use dry chamber data or real dive in cold water data for instance would yield different results if the same dive profile results in drastically different bubbles counts in those conditions. There are many interconnected factors that could explain the differences observed, including temperature, immersion, exercise, hydration, but also individual fitness. The wet/dry differences can be due in part to temperature differences since diving with a wet suit would result in colder dive conditions. Hemodynamic distribution within the vasculature could also be at play since the microgravity effect of water submersion would result in a redistribution of blood volume [147]. In cold conditions this would be combined with vasoconstriction in the extremities to some degree [11]. Systematic studies looking at bubble counts in wet and dry diving whilst closely matching other conditions would be useful. For instance a study looking systematically at the same square dive profile could be devised for in water dives with drysuits and wetsuit exposures, and also in chamber conditions. The temperature and exercise conditions would need to match. Another factor as far as water diving is concerned could be the diver's orientation. In astronauts there is a known adaptive mechanism which happens due to the shift of fluids in the upper part of the body. A diver which would ascend on a line could be in upright position. In technical diving where decompression procedures can last for hours a diver is not likely to stay horizontal for the whole duration of the dive. To the best of our knowledge no study has been performed where the diver's position has been systematically investigated, either in chamber or water conditions.

2.1.4 Conclusion

Nanobubbles spontaneously forming on hydrophobic surfaces, observed via atomic force microscopy, constitute a potential candidate for micronuclei, although their capacity for growth is still debated as they are very stable. Heterogeneous nucleation and tribonucleation therefore still hold as the prime candidates for bubble formation in human hyperbaric exposures, homogeneous nucleation needing far greater pressure differences than those encountered. Stabilization processes for micronuclei have been revised and some new ones proposed. Hydrophobicity of surfaces seems to be an important factor in crevice growth 67 models and could potentially relate to some physiological studies where adiposity as a risk factor has been investigated. An alternative stabilization mechanism is that of stabilizing potential wells due to tissue elasticity, which combined with continuous degrees of tribonucleation from body movement could permit a constant supply of micronuclei and potentially explain some of the physiological studies on the role of exercise in bubble formation. In any case, incorporating bubble formation and growth mechanisms in decompression models is important and the general direction of research in that area is an effort to make models more biophysical to allow better extrapolation. In that respect a consistent quantitative venous bubble monitoring system post dive, unambiguous and reliable needs to be developed to calibrate and verify these results. The location of micronuclei or where bubbles form remains unanswered, with tissue bubbles (in situ) now having been presumably observed in vivo in addition to circulating bubbles [139, 148]. The interaction between them, as well as between multiple bubbles has been shown to result in dissolved gas competition for growth where flow conditions and perfusion rates are dominant parameters. Furthermore the single gas plug possibility was shown to be worthy of more careful consideration as far as DCS is concerned. Finally, a closely controlled study looking at wet/dry dive differences would be useful to evaluate how many of the differences observed can be accounted for by other related parameters such as temperature and exercise.

2.2 Circulatory bubble dynamics: from physical to biological 6 aspects

2.2.1 Aim and scope

The physics of bubble growth and detachment is routinely discussed with respect to engineering applications, for example to design optimal pipes for oil transport, propeller design, flotation devices etc. [149-152]. There are however also instances where bubbles form or are introduced in vivo. In these cases the bubbles can be found in the bloodstream or in tissues. Bubble growth and detachment physics then becomes significant in predicting and controlling the probability of these bubbles causing mechanical problems by blocking vessels, displacing tissues, or inducing an inflammatory cascade if they persist for too long in the body before being dissolved.

These bubbles in the bloodstream may result in ischemic problems from direct vessel occlusion, which can be fatal in the arterial side of the circulation. Bubbles in the vasculature can obstruct blood flow, activate inflammatory pathways and cause clotting [153]. The interaction of the bubbles with the components of the blood is complex. It has been shown for instance that an adhesion force causing bubbles to lodge in the vasculature is created by the interaction between the macromolecules in the blood and the bubble which pushes it towards the endothelium [154].

The review in this Section 2.2 presents the mathematical formalism related to the physics of bubble growth from decompression and attempts to discuss it in the context of bubbles in the human bloodstream. Modelling bubble growth in the circulation can help determine the probability of them blocking the circulation directly in a given region. In addition, it is important for the determination of their persistence and rheology in the circulation, also related to the inflammation cascade, by modelling the introduced microbubble flow rate in the mixed venous blood from formation sites.

6 Adapted from V. Papadopoulou et al. Adv Colloid Interface Sci.,2014; 206:239-249. 69

Most work in the field of decompression induced bubble growth has not been done in the in vivo context of interest here (human bloodstream), but primarily in the fields of geology (volcanic eruptions, magmas, etc.) and industrial applications (multiphase dynamics, bubbly flows in pipes, etc.) [155-159]. In addition the exact composition of the bubbles that form in vivo due to hyperbaric decompression and initial site of formation of those bubbles remain somewhat open questions as seen in Section 2.1 [160].

By contrast to decompression induced bubbles whose site of initial formation and exact composition are debated, there are other instances of bubbles in the bloodstream which are well-defined. These are gas emboli unwillingly introduced during surgical procedures and ultrasound microbubbles injected for use as contrast or drug delivery agents. We will therefore present the general bubble growth formalism and its inertially controlled growth case in particular, before classifying the analytical or for the most part numerical solutions of the key equations with respect to the different simplifications used. Blood is then discussed as a flowing liquid where bubbles grow and detach. We finish by looking into what we can learn from the behaviour of bubbles in the human bloodstream, specifically bubble dissolution, rheology and biological interactions for the different cases of bubble and blood composition considered.

2.2.2 Background: how can bubbles end up in the blood stream?

Bubbles introduced in the circulation can have iatrogenic causes: either as a side effect of vascular and heart surgery, injections and skin transplants or on purpose in the case of embolotherapy or ultrasound contrast agents where bubbles are used to respectively deliver drugs to targeted sites or image the vasculature. Furthermore endogenous bubbles can also form in the body due to mechanical heart valves or during or after decompression from pressure exposures such as those undergone by scuba divers and astronauts and are therefore an occupational hazard for caisson and tunnel workers as well. Table 1 presents a brief summary of the characteristics of these different bubbles.

Persistence in the Approx. diameter (μm) Gas content Shell properties bloodstream

Debated. The detectable VGE size using Observed circulating for up linear B-mode ultrasound is Inert gas breathed to 3hours post-dive using Decompression above 20-30μm [161, 162]. Unknown, would depend mainly (nitrogen ultrasound imaging bubble Theoretical predictions on formation site usually) (supersaturated tissues calculate bubbles of more than state) [160] 5-10μm [163]. Animal studies have observed 19-700μm [164].

Gas emboli Ambient air of surgery Highly variable, 15-100μm, from surgery room or anesthetic gas Unencapsulated Depends on severity slug form often observed [165-167] used

Microbubbles remain for Contrast agents Encapsulated (lipid or several minutes in the for ultrasound Usually heavy gas protein). bloodstream enhancing and Drug 1-7μm such as Can have functionalized contrast, but bubbles can delivery perfluorocarbon shell to attach to particular remain in the body longer bubbles [168] sites or carry drugs/DNA (accumulate in spleen and liver over 30min)

Table 1 : Comparison overview between circulatory bubble types found in the bloodstream.

2.2.2.1 Bubbles introduced in the bloodstream on purpose

Ultrasound imaging has the advantage of being a non-invasive, non-ionizing and low-cost imaging modality. It uses sound waves (typically between 1 and 15 MHz high-frequency pressure waves) to visualize organs and blood flow. However compared to other imaging modalities such as MRI or CT, the images produced are sometimes of lower quality due to excessive attenuation (absorption, scattering and reflection) and distortion from the reflectivity of tissue boundaries. In addition, red blood cells scatter ultrasound poorly due to their relative acoustic impedance mismatch to plasma [169] resulting in the blood appearing dark on B-mode images [170]. Many approaches and new techniques are being developed to overcome some of the challenges related to ultrasound imaging, one of the most important being the development of stabilized gas microbubbles injected intravenously for enhancing the echogenicity from within the blood (contrast agent) [171]. These microbubbles of diameter less than 10 μm oscillate radially when hit by the pressure waves from ultrasound imaging and scatter those in all directions, thus increasing the contrast of the image drastically by effectively acting as secondary “point” sources of sound themselves (Figure 7). Their scattering power is very big due to the large impedance difference between their gas core and the surrounding liquid and they can be driven at their resonant frequency for big

71 radial amplitude oscillations. These microbubbles are encapsulated to last longer in the blood circulation, as free bubbles dissolve quickly [171], and so that they do not coalesce (e.g. PEGs on the shell). Acoustic signals from ultrasound contrast agents can be separated from those of tissue due to the fact that bubbles' non-linear response is much higher than that of soft tissues [172], making perfusion imaging possible.

Figure 7 : Ultrasound imaging of a human liver with microbubble contrast agents: non- linear mode preferentially showing bubbles (left) and standard linear B-mode image (right)

The encapsulating layer of the microbubbles (protein or lipid) can be bound to molecules that attach to specific target sites in the body, for instance antibody, receptor or ligand of a cancer cell type or inflamed endothelium, etc. (Figure 8). By functionalizing their shell, microbubbles can therefore be used for molecular imaging. In addition, microbubbles can be loaded with therapeutic agents or with DNA and act respectively as drug or gene delivery agents to specific target sites [173]. Primary and secondary ultrasound radiation forces, due to the incident pressure field and scattering by resonating bubbles respectively, can be used to push injected microbubbles towards their target sites and increase their chances of binding. It has also been demonstrated that the oscillation of bubbles next to cells under a focused ultrasound field can open up cell membranes for delivery of therapeutic agents [174].

Figure 8 : Diagram of targeted microbubbles shell composition [3]

2.2.2.2 Unwanted bubbles in the bloodstream

Bubbles can form from micronuclei as a result of ultrasonic cavitation at high exposures [175]. Bubbles have also been shown to be created by cavitation where prosthetic heart valves move by disturbance of the normal flow at closure [176, 177], with instances of up to 620 embolic events over 30 min recorded in patients [166]. Another iatrogenic cause of gas embolism is from surgery, where air is introduced in the vasculature during injections, catheter placing, etc. These bubbles can then be associated with all the serious clinical complications related to gas embolism, including ischemia, stroke and cardiac failure. Iatrogenic gas embolism is estimated to occur in only 2.65 per 100 000 hospitalizations, but with high long-term mortality and morbidity [178]. Gas embolism as a result of cardiopulmonary bypass for instance is estimated to result in cognitive decline consistent with cerebrovascular embolization restricting blood flow into localized areas of the brain for 25% of patients [179-181]. Seventy to eighty percent of strokes, the second leading cause of death worldwide [182], are ischemic in origin, resulting directly from cerebral thrombosis or embolism blocking an artery in the brain or in the neck [183].

In addition, microbubbles can grow in the body due to changes in ambient pressure (scuba diving, extravehicular excursions for astronauts, etc.). The precise formation mechanism and site of these bubbles are still debated and the subject of Section 2.1 [160]: potential stabilizing mechanisms for micronuclei from which bubbles can grow are hydrophobicity of surfaces and tissue elasticity, with formation sites in facilitating regions with surfactants, hydrophobic surfaces or crevices, such as caveolae or between endothelial cells.

Intravascular bubbles post pressure-excursions are routinely observed with Doppler and B- mode ultrasound imaging. Human blood and tissue contain dissolved inert gases from respiration and cell metabolism, the molar concentrations of which are proportional to the gas partial pressure (Henry's law). The partial pressure of inert gases breathed is increased during a scuba diving descent, where the diver is supplied with air at ambient pressure throughout the dive (the ambient pressure at depth increases by roughly 1 atm for every 10 meter depth). This results in inert gases, not utilized by the body, being absorbed by tissues and blood during the descent phase of the dive. Metabolic gases, oxygen bound for the most part by hemoglobin, and carbon dioxide, do not cause problems as they are directly used by the body and recycled through breathing. Bubbles can grow during or after the decompression phase of the dive (ascent), when the inert gas (nitrogen and/or helium usually) that has been dissolved in the tissues during the compression phase (descent), is released in the circulation (Figure 9). The tissues are then supersaturated and release inert gases in the form of in situ tissue bubbles or bubbles that form in, or enter, the bloodstream.

Circulating bubbles are normally filtered out by the lungs (expiration) [184] and do not pass into the arterial side of the circulation, provided that they are big enough to be trapped and dissolved but small enough not to obstruct any vessels upstream of the lungs, and that their number does not impair the lung capability to filter them out. Failure to ascend slowly enough (following decompression stops where the diver waits at certain depths before resuming his ascent) to control the number and size of those bubbles can result in potentially fatal decompression sickness. In addition, the presence of venous-arterial shunts (cardiac Patent Foramen Ovale – PFO [148, 185], lung shunts – IPAVA [186], …) can provide paradoxical entry for these bubbles into the arterial side of the circulation [185]. This effectively results in arterial gas embolism with the added complication of the body being more saturated in inert gas, which hinders bubble dissolution.

Figure 9 : Ultrasound imaging of the heart 1 hour post scuba dive. Venous gas emboli (VGE) are circulating in the right heart chambers (delimited in white, image reversed). The lungs are effectively filtering these VGE and they do not appear in the left heart chambers.

2.2.2.3 Life cycle of a bubble in blood

Figure 10 : Schematic of bubbles from decompression in the circulation (inverted orientation in keeping with echocardiography frames)

Once a bubble has entered the bloodstream, let's assume on the venous side, if its size is small enough not to directly occlude a vessel it will be transported by the blood and follow the normal circulation into the right heart and then the lungs. The bubble will travel at the velocity of blood provided it is small compared to the vessel cross-sectional area, with velocities from 0.03 cm/s in the capillaries to 40 cm/s in the aorta and 15 cm/s in the vena cavae [187]. Microbubbles arriving in the lung capillaries are normally trapped if big enough, then dissolved during expiration [184].

In addition to a direct entry into the arterial side of the circulation, bubbles originally from the venous side can gain paradoxical entry into the arterial circulation in the presence of cardiac or pulmonary shunts (Figure 10). A patent foramen ovale or PFO is one such venous- arterial shunt located in the atrial septum of the heart, with estimated prevalence in the population of 25–33% [148, 185]. Another is intrapulmonary arterial-venous anastomoses (IPAVA) which allow blood to bypass the pulmonary microcirculation [186]. Both of these shunting mechanisms have been shown to be exacerbated by exercise [186] and provoking mechanisms that increase the intra-thoracic pressure [148, 185].

Finally, bubbles can also enter the arterial circulation if the lung does not act as an efficient filter. It was shown in dog experiments that this can happen if the bubbles are too small to get trapped (less than 22 μm [184]) or if there are too many bubbles causing a deformation or rupture of the lung capillaries (administration of single bolus over 20 ml of gas [184] or 0.15 ml/kg·min for over 30 min [188]). Once a bubble is in the arterial side of the circulation the probability of it blocking the circulation increases as, conversely to the venous side of the circulation, the blood is transported into progressively smaller vessels and capillaries. This is especially the case if the bubbles continue to grow by merging into one another or due to the continued ascent and degassing in scuba diving.

Depending on the saturation dynamics of the rest of the blood and tissues, the bubble may dissolve or expand accordingly during transport. In any case the inflammation cascade initiates as the bubble persists in the circulation and gets in contact with other blood cells suspended in plasma or brushes against endothelial cells [189]. There is also evidence of macrophages internalizing bubbles and bubbles accumulating in the liver and spleen before the gas eventually diffuses out [190]. This is described in more detail in Section 2.2.4.4 Biological interactions.

2.2.3 Bubble growth and detachment from decompression

Notwithstanding chemical interactions, a supersaturated state can be reached in a liquid solvent containing dissolved gas as a result of a reduction in ambient pressure or rise in liquid temperature. Above a respective critical supersaturation level, gas bubbles will nucleate heterogeneously or homogenously. The rate of gas desorption from the liquid therefore is associated with both the nucleation and the subsequent growth rates of those bubbles once formed, as well as their detachment rate. While bubble growth is primarily linked to gas mass transfer from the bulk liquid to a gas nucleus, bubble detachment is dictated by buoyancy and/or flow induced shear forces which both act to destabilize the bubble contact with the solid surface beneath and drag it away from its nucleation site [191].

Bubble growth in the bloodstream can be induced by different mechanisms related to heat and/or pressure. These range from the flow conditions in the circulation to the oscillations induced by ultrasound imaging of bubbles (changing pressure and heat of contrast agent bubbles). In the case of interest here we consider the ambient pressure changes as the main driving force of growth: a rapid or gradual, staged or continuous, pressure decrease (decompression) in solutions with dissolved gas as an analogue to the case of the bubbles observed in blood in scuba divers for instance.

2.2.3.1 General formalism: heat and mass transfer

A small gas bubble which has nucleated on a solid surface in contact with a liquid containing this dissolved gas can grow and finally detach from it [119]. The description of the growth rate of the bubble requires the coupling of the equations of motion, continuity, conservation of diffusing species and heat transfer, and must account for convection, viscous and surface tension forces, as well as other flow conditions [115, 191, 192].

In the simple case with no flow (basin degassing) at constant temperature, after the initial growth phase, the growth of the bubble will be theoretically governed by molecular diffusion with a scaling relationship of the form 푅 ~ 푡푎, where R is the bubble radius, t the time of growth and a the scaling factor [116, 117]. However the range of scaling factors 푎 was obtained experimentally and values differ not only between classes of problems, but also for the same type of experiments [117, 191, 192]. The difficulty comes from devising

77 experiments to test the theoretical predictions, as it is very difficult to decouple heat and mass transfer in experiments, for instance to get a decompression driven bubble growth at constant temperature.

In general bubble growth can be separated into two categories depending on the driving force [193]: bubble growth that is primarily controlled either thermally [115, 194, 195] or inertially (mass transfer) [196] such as in the decompression case.

We review below the basic considerations for the heat diffusion case, before treating in more detail the bubble growth resulting from decompression of a liquid containing dissolved gas in Section 2.2.3.3.

2.2.3.2 Thermal degassing (heat transfer controlled)

Scriven [195] derived the influence of radial convection on bubble growth from molecular diffusion of the gas molecules through the liquid and the liquid–gas interface. The outward convective movement of the growing bubble front is shown to result in an effective higher molecular diffusion and thus an increase in growth rate than that expected from diffusion considerations alone:

R = 2b√kt, (3)

where 푅 is the bubble radius, 푡 growth time, 푏 a dimensionless growth constant and k a diffusion term. In the case of heat conduction governed growth, 푘 is the thermal diffusivity, and for growth governed my molecular diffusion 푘 is the diffusivity. Equation 4 shows that the radius is proportional to the square root of time for a negligible Laplace pressure. The bubble growth constant 푏 was also explicitly related to the densities and concentrations of gas and liquid through a dimensionless superheat or supersaturation parameter

ρ ∙(C −C ) φ = φ{ε, b} = L 0 sat , (4) ρG∙(ρL−Csat) with

ρ ε = 1 − G, (5) ρL

where 휌퐺 and 휌퐿 are the densities of the gas and liquid respectively, and 퐶0 and 퐶푠푎푡 the mass based concentrations of the bulk and equilibrium concentrations of the gas in the liquid.

The above 1D analysis has been extended to include temperature dependent transport properties of the liquid and the gas as well as a 2D description of bubble growth and motion on a non-flat surface [117, 191].

2.2.3.3 Decompression degassing (mass transfer controlled)

2.2.3.3.1. Pool degassing (diffusion controlled)

The desorption or release of the dissolved gas in a fluid brought to supersaturation can be done either solely by diffusion through the liquid contact area to the air for instance or in a nucleating fashion, in other words through the formation of bubbles. In the latter case, desorption is not simply the analogue of the absorption kinetics and the diffusive mass transfer equations are significantly altered by the hydrodynamic conditions in the liquid.

Desorption without bubble creation has been examined [197, 198], however the case of desorption with bubble formation has rarely been considered [199]. For these, a first factor must be the availability of nucleation sites for the bubbles to form due to small pressure ratio reductions. For quiescent solutions (no flow), the conditions under which the bubbling degassing mechanism dominates over normal diffusive mechanisms were first investigated by Schweitzer et al. [200] and Burrows et al. [201], showing that bubbles could be prevented if the pressure reduction rate was slow enough for total pressure reduction ratios of up to 3.75. It is, of course, expected that if the ambient pressure is below the sum of partial pressures of the solute and solvent, then under isothermal conditions at constant volume the system will tend to reinstate pressure equalities by gas and vapor evolution [198].

Isothermal bubble growth from capillary effects, inertia, viscosity and mass diffusion of a solute gas from a liquid was investigated theoretically by Langlois [202]. Neglecting the inertia of the liquid, the radius of the bubble was shown to satisfy

3 4μL dR 2σL pg0R0 + = AC + − P , (6) R dt R sat R3 amb

79 where 푅 is the bubble radius, 푅0 the initial bubble radius, 휇퐿 the viscosity of the liquid, 휎퐿 the surface tension of the liquid, 푃푎푚푏 the ambient pressure, 푝푔0 the initial partial pressure and

퐶푠푎푡 the concentration of solute gas related by Boyle's law to its partial pressure such that

푝푔 = 퐴퐶푠푎푡 with 퐴 being a constant.

Once formed, the general growth and dissolution of microbubbles in a solution from diffusion considerations was derived by Epstein and Plesset [92]:

dR DGBT⁄MG(C0−Csat) 1 1 = ( + ) ( 7 ) dt Pamb+4σB⁄3R R √πDgt

where 푅 is the radius of the bubble, 푡 time, 퐷퐺 the gas mass diffusivity in the fluid, 퐵 the universal gas constant, 푀퐺 the molecular weight of the gas, 퐶0 the bulk dissolved gas concentration, 퐶푠푎푡 the saturated concentration of the dissolved gas, 푃푎푚푏 the ambient pressure, 휎퐵 the surface tension of the bubble and 푇 the temperature. Equation 7 predicts that the bubble will dissolve for saturated solutions (퐶0 = 퐶푠푎푡) if its surface tension 휎퐵 ≠ 0.

Equation 7 predicts a dissolution time for a perfectly spherical air bubble in pure water of 1 s for a 1 μm radius bubble, 1–6 s for a 10 μm bubble and 11–70 days for a 1 mm bubble [203]. However bubbles can take a cylindrical form in vivo, in which case the dissolution time for a bubble of same total volume is multiplied by a factor of at least 2 [203].

The Epstein–Plesset model was most comprehensively tested for the effects of surface tension and undersaturation recently using a new micromanipulation (micropipette) technique [204]. Surface tension was studied using single and double-chain surfactants and undersaturation coating the microbubble with a wax monolayer from solid distearoylphosphocholine lipid to effectively achieve a zero surface tension. The model was shown to underpredict the resolution time of microbubbles due to surface tension in the range of 72–25 nN/m by an average of 8.6% and overpredict this time due to undersaturation by 8.2% in the range of gas saturation from 70 to 100%.

2.2.3.3.2. Flow degassing (inertia controlled)

Fluid mechanics also play a role in the conditions for maintaining bubbling and we can consider either a basin/pool degassing scenario as treated above or a flowing liquid degassing for which calculations of desorption rates by mass transfer are even more limited [205].

The rate of gas desorption with bubble formation from agitated liquids was investigated using a stirred cell apparatus from CO2 supersaturated aqueous solutions [198]. It was found that in a flow system considerable bubbling does not start until the partial pressure of the dissolved gas alone is greater than the ambient pressure, and only a minimal amount of bubble evolution occurs when the sum of partial pressures of the components of the liquid exceed the total pressure.

The same flat gas–liquid interface system was then studied in a continuously baffled agitated vessel and the rates of desorption of carbon dioxide and nitrogen from supersaturated water solutions measured at different temperatures to extract volumetric mass transfer coefficients for desorption and correlate those to the relative supersaturation of the solution and its Reynolds number [206].

In a more recent study, the volumetric mass transfer coefficients for the bubbling desorption of CO2 from DMEPEG solutions were correlated by a power relationship to supersaturation, Reynolds and Weber numbers [205]. It therefore appears that the viscosity and surface tension of the fluid with respect to its inertia can be important quantities in addition to the degree of supersaturation.

2.2.3.4 Bubble detachment

2.2.3.4.1. In stagnant liquid

Let us consider the forces acting upon a bubble growing on a surface in a liquid. The forces that contribute to its adhesion to the surface are the surface tension 퐹푆 and drag produced by the bubble growth 퐹퐷, whereas the forces which pull it to detach are buoyancy 퐹퐵, pressure

퐹푃 and liquid inertial 퐹퐼 forces [192]. So the force balance equation is given by:

FS + FD = FB + FP + FI, ( 8 )

81 which, for slow growth rates, simplifies to

FS = FB + FP. ( 9 )

The time of detachment is a function of the contact angle of the bubble with the surface, the surface geometry itself, as well as the flow conditions. For bubbles growing from conical cavities, where the contact angle of the growing bubble is 휃, at detachment the surface adhesion force is just balanced by the pressure and buoyancy forces, such that

2γ 2πR σ sin θ = ( − ΔρgH) πR2 + ΔρσV, ( 10 ) d R′ d

where 푅푑 is the radius of the bubble at detachment, 퐻 the bubble height, 푅′ the radius of curvature of the bubble at its highest point, 푔 the acceleration due to gravity, 훥휌 the mass density difference between the gas of the bubble and the fluid (훥휌 ≡ 휌퐺 − 휌퐿), 휎 the interfacial tension between the fluid and the gas, and 푉 the bubble volume.

Chappell et al. investigated the effect of the cavity geometry on growth rate and detachment [207, 208], showing that once the bubble has emerged from the cavity its behaviour is determined by the size of the opening. The flow conditions, not considered explicitly in this cavity geometry effect modelling, are expected to dictate the behaviour for detachment much more than the cavity geometry after the initial bubble growth phase [61].

2.2.3.4.2. In flowing liquid

The ensemble of forces pulling the bubble upwards or lift force experienced by a bubble moving in the liquid flow can be much more complicated. For example in a non-symmetrical flow, the bubble experiences a lift force perpendicular to its plane of motion and many ad hoc expressions of the lift force are used in bubbly flows to reconcile the experimental observations [209]. Experiments in microgravity showed that bubbles tend to migrate towards the centre of the pipe [210].

Figure 11 : Geometrical definitions of parameters for the velocities and force balance equations [211]

The bubble detachment criterion of a bubble growing at the wall of a linear shear viscous flow can be obtained from a force balance equation derived by Duhar et al. [211, 212]:

Fc + FCP + FB − ∫ (PL − PL Z0)퐧 dS + ∫ τL퐧 dS = ퟎ, ( 11 ) SB SB

with the capillary force 퐹퐶 such that [213]

π(α − β) F = 1.25 × 2 r σ (sin α + sin β)퐞 C C L π2 − (α − β)2 퐗 (12) π − 2 r σ (cos β − cos α) 퐞 C L (α − β) 퐙 the contact pressure 푭푪푷 2휎 퐹 = ( 퐿 − 휌 푔퐻) 휋푟2풆 , (13) 퐶푃 푅 퐿 퐶 풁

and the buoyancy force 퐹퐵 encompassing both gravity and Archimedes forces

4 퐹 = (휌 − 휌 ) 휋푅3품 (14) 퐵 퐺 퐿 3

where 푆퐵 is the surface of the bubble, 풏 the unit vector normal to the bubble surface, 품 the gravitational acceleration, 휌퐺 the bubble gas density, 휌퐿 the density of the liquid, 푃퐿 the pressure of the surrounding liquid, 푃퐿푍0 the reference pressure in the liquid at the wall (at

푍 = 0 with respect to Figure 11), 푅 the bubble radius, 휎퐿 the surface tension of the liquid, 푟퐶 83 the orifice radius of the conical cavity from which the bubble grows, 퐻 the bubble height, 푎 the advancing contact angle, 훽 the receding contact angle, 풆푿 the unit vector parallel to the wall, 풆풁 the unit vector perpendicular to the wall and 흉푳 the deviatoric stress tensor due to viscous effects.

The first 3 terms of equation 11 are the static forces and the last 2 terms the hydrodynamic forces encompassing the drag, migration and unsteady forces. The contact angles (Figure 11) were shown to continuously evolve during growth and detachment occurs for maximum capillary force value of the advancing contact angle, 70° for a quiescent liquid and 90° in a shear flow. In addition, Duhar et al. [211] derived simplified departure radius equations (less than 5% error) for both the quiescent liquid and shear flow cases.

It remains however to be determined to which extent the above formulations apply to the case of bubbles in the bloodstream. The lack of information in this regard is no doubt due to the practical difficulty of observing bubble detachment from vessels in vivo, which would in addition require creating these bubbles and predicting their nucleation site. A first step in this direction was the recent demonstration ex-vivo of bubble formation on the surface of hydrophobic vessels [214].

Chappell et al. [215] modelled bubble detachment in blood for bubbles growing in cavities of the vessels' walls during decompression from a saturation dive. The detachment process was separated into two distinct phases: bubble deformation with the bubble growing deflected sideways in the direction of the flow but still attached to the cavity mouth and assumed approximately spherical, then detachment once the bubble separated from the cavity. The detachment happens when the drag force experienced by the bubble due to the blood flow exceeds the capillary force. Assuming that the flow around the spherical bubble is laminar (Reynolds number ~0.001) and that the variation ψ of contact angles 푎 and 훽 is symmetrical a force balance between the drag and capillary forces gives

2 6휇 푅′ 휓 = 퐿 ( ) , (15) 휎퐿 푟퐶

where 휇퐿 is the blood viscosity, 휎퐿 the surface tension, 푟퐶 the radius of the crevice mouth and

푅′ the radius of curvature of the bubble. The limiting value of deformation angle 휓푐푟푖푡 is then

84 used to get the non-dimensional radius of the bubble at the end of the deformation phase,

푅′푐푟푖푡/푅, by rearranging equation 15 to get 푅′푐푟푖푡.

2.2.4 Bubble behaviour in the bloodstream

2.2.4.1 Bubble dissolution in blood

In the case of decompression bubbles growing in tissue or blood, assumed liquid for simplicity, the nucleation and subsequent growth of decompression bubbles is dependent on the dissolved gas content (concentration and diffusivity). The solubility and mass diffusivity of different inert gases (commonly nitrogen and helium in scuba diving) are not identical and influence bubble dynamics. Experimental values for diffusion and solubility coefficients for gases in biological tissues and fluids are scarce and reviewed in [216]. It is estimated [216] that diffusion coefficients in tissues are 25% to 50% lower than for water and that the solubility coefficients for water can be used as an approximation (less than 20% error) for most tissues apart from fat tissues which have higher coefficients.

Once formed, a gas nuclei (bubble of at least the critical radius size to avoid dissolution due to surface tension effects) will tend to equilibrate the differences between the dissolved gas tension in the liquid and the bubble gas pressure at its interface.

The Epstein–Plesset (eq. 7) presented in Section 3.3.1 was adapted for multi-gas-component bubbles without encapsulation in blood by Kabalnov et al. [217]:

푑푅 푝푒푥푐+2휎퐵/푅 1 1 = −퐷퐺퐿퐺 ( + ), (16) 푑푡 푃푎푚푏+4휎퐵⁄3푅 푅 √휋퐷퐺푡

where 퐷퐺 is the gas diffusivity, 퐿퐺 the partition coefficient of the gas between the liquid phase and the bubble and 푝푒푥푐 the excess pressure (sum of excess systemic blood pressure with excess partial pressure of the dissolved gases from the atmosphere to the bloodstream).

The evolution of bubbles in supersaturated tissue is particularly relevant to DCS and was investigated theoretically by Srinivasan et al. [218, 219] using a modification of the Epstein

85 and Plesset model with three regions (bubble inside tissue mass but separated from it by an unstirred boundary layer of constant thickness).

The detachment model developed by Chappell et al. [215] calculated the possibility of a gas plug directly occluding the capillary in which the bubble forms after the dive from the distance travelled by the bubble laterally, assuming the bubble travels at the same speed as blood flow velocity. The gas concentration in the fluid (blood) was assumed to change much slower than the gas transfers into the bubble, with a boundary layer of finite thickness and varying gas concentration around the bubble. It was shown that a bubble growing after a dive could theoretically create a gas plug directly in the capillary where it originated, although vessel deformation and interaction of the bubble with the flow field were not considered.

The physiological concept of the oxygen window, or inherent unsaturation due to oxygen metabolism [220], was used by Van Liew and Burkard [221] to explain how stabilized microbubbles can persist in the bloodstream. This was done simply by equating the pressure on the bubble surface to the sum of the partial pressures of gases inside the bubble. They also suggested that gases inside the bubble can be at diffusion equilibrium if the bubble is able to support some degree of negative pressure mechanically.

Looking at bubble dissolution, the extravascular bubble model by Van Liew [222] was applied to the circulating bubble [223]. The supersaturation state of the mixed venous blood during decompression means that the bubbles will dissolve slower or even expand, potentially resulting in a lung filter overload.

A mathematical model to account for the cylindrical geometry of bubbles observed in vivo was also developed [224], showing that the absorption time observed corresponded better to the predictions of the model. Once the length of the cylindrical bubble has shrunk, it reduces to a spherical geometry and any subsequent shrinkage will decrease its diameter. This geometry results in longer absorption times than expected for spherical bubbles, as observed with gas emboli in vivo [224, 225]. A similar model using this in vivo bubble geometry in vessels was developed for use in hyperbaric oxygen treatment, to determine the most effective treatment protocols for cerebral gas embolism [226].

Finally, the intravascular exogenous surfactant concentration was not found to influence initial bubble conformation, but increased bubble breakup and the rate of bubble reabsorption [227]. In an in vivo rat model of cerebrovascular arterial gas embolism, intravenously 86 injecting them with a surfactant prior to inducing cerebral gas embolism showed a prophylactic effect: strokes were undetectable on brain MRI scans and post embolic cognitive and sensorimotor deficits were significantly reduced [228].

2.2.4.2 Bubble dynamics in an ultrasound field

The equations for microbubbles in an ultrasound field are mostly modifications of the Rayleigh–Plesset equation [229, 230] describing the dynamics of a free perfectly spherical bubble surrounded by an unbounded viscous incompressible liquid in the far-field 푃∞(푡) at constant liquid temperature and uniform pressure in the bubble 푃퐵(푅):

푑2푅 3 푑푅 2 4휇 푑푅 2휎 휌 [푅 + ( ) ] + 퐿 + 퐵 = 푃 (푅) − 푃 (푡), (17) 퐿 푑푡2 2 푑푡 푅 푑푡 푅 퐵 ∞

where 휇퐿 and 휌퐿 are respectively the viscosity and density of the liquid and 휎퐵 the surface tension of the bubble of radius 푅.

Assuming that the vapour pressure in the bubble and non-linear terms are negligible and that the pressure field far from the bubble changes sinusoidally [230],

3훾 푅0 2휎퐵 푃퐵(푅) = 푃퐺(푅) = 푃퐺 푒푞 ( ) ; 푃퐺 푒푞 = 푃푒푞 + ; 푅 푅0 (18)

푃∞(푡) = 푃푒푞 − 푃퐴 푠푖푛(훺푡),

the radial oscillatory behaviour of the bubbles can be derived by approximating a first-order solution to this equation as

퐴 (훺) 푅 ≈ 푅 (푡) = 1푟 푃 푒푥푝 (푖훺푡), ( 19 ) 1 2 퐴

where the subscript 푒푞 refers to the quantity at equilibrium, 푃퐴 is the pressure amplitude and

퐴1푟(훺) the linear amplitude–frequency response of the bubble which depends on the liquid density and viscosity, the gas pressure (or sum of partial pressures of gases) in the bubble 푃퐺, the adiabatic exponent 훾 (ratio of the heat capacity at constant pressure to heat capacity at constant volume for the gas mixture) and the angular transmitted frequency 훺 = 2휋푓 with 푓 87 the frequency of bubble oscillation. The full derivation and explicit equations for the linear amplitude–frequency of the bubble response 퐴1푟(훺) can be found in [72, 231].

Some of the effects that are not taken into account in equation 17 are the heat conduction and gas diffusion through the bubble wall, the blood composition making compressibility of the fluid and viscoelasticity effects potentially important, the shell effect, as well as the interactions between the bubbles and the blood cells and endothelial boundaries.

For small radial oscillations, corrections to equation 17 have been developed to this effect [232, 233], which led to the de Jong models assuming that the shell of the bubbles dominates their response [234, 235]. However Sboros et al. [236] showed that the behaviour of those bubbles is not compatible with the theoretical predictions of the viscoelastic ball model of de Jong. Another theoretical model for an encapsulated bubble was suggested by Church [237] who considered a finite thickness for the shell, separating the interface of the bubble into two layers of different surface tension to reflect the different boundaries with gas and liquid respectively. In addition, a complicating factor of the modelling is the fact that oscillating bubbles do not always remain spherical in shape [167] and are clearly asymmetric near the vessel walls [238].

An important consequence of equation 17 is that over time the oscillating bubble near resonance can be growing due to rectified diffusion [239]. The diffusion rate of gas into the bubble is indeed proportional to its surface area and therefore the net effect of oscillations will be to grow the bubble over each oscillation cycle. It has been shown that the effect of rectified diffusion is accelerated in the presence of surfactants at the bubble-liquid boundary [165].

2.2.4.3 Rheology of microbubbles in the bloodstream

2.2.4.3.1. Brief overview of blood rheology

Blood plasma can be approximated to a Newtonian liquid [240] in the case of studies involving microbubbles in arteries or in vitro conditions in an ultrasound field. This is because viscoelastic effects (due to blood cell deformation and vessel walls) can be ignored if the bubble is growing more than 25 bubble radii away from the vessel walls [241]. However

88 for bubbles in small veins and capillaries, as well as those growing in tissue, this is not the case and those cannot be modelled as simple Newtonian fluids [242].

Blood is a mixture of suspended red blood cells, white blood cells and platelets in plasma. Plasma contains mainly water and proteins and lipids, and is a Newtonian fluid, i.e. its shear stress is linear with the strain rate [243]. The non-Newtonian rheology exhibited by blood is down to the cellular components. When these are small compared to the diameter of the vessel, this effect can be ignored and blood is approximated to a continuum. For an incompressible Newtonian fluid, the conservation of fluid mass and that of linear momentum yield the following equations respectively [244]:

훻 ⋅ 풖 = 0, (20)

and 휕풖 휌 ( + (풖 ⋅ 훻)풖) = −훻푃 + 휇 훻2풖 + 휌 품, ( 21 ) 퐿 휕푡 퐿 퐿 where 휌퐿 is the fluid density, 품 the acceleration due to gravity, 풖 the velocity vector, 푃 the pressure, and 휇퐿 the fluid viscosity. For non-Newtonian fluids, the shear stress term in equation 21 is changed to reflect the shear-stress relationship for that fluid.

Over a range of shear stresses which depend on its haematocrit content, blood is a shear thinning fluid, i.e. its viscosity decreases with an increasing strain rate. Below a critical yield stress value of shear stress [243], blood does not behave like a viscoelastic fluid anymore, but like a solid. A constant effective fluid viscosity can be implemented, or the non-Newtonian properties of blood can be approximated by the Casson model [243]. For blood flow in the capillaries, the diameter of the red blood cells is similar to that of the vessel and they have to deform and go through one at a time in a file, so the continuum approximation is not valid [245-247].

Equations 20 and 21 are subject to boundary conditions to account for the circulation of blood from one region of flow to the other (continuum) and also for the wall condition (elasticity etc.). Blood flow in the arteries is generally pulsatile, whereas it is nearly steady in capillaries and veins [248, 249].

It is therefore clear from the brief overview above that different forces resulting from these flow conditions can act upon the bubbles, thus giving different bubble behaviour.

2.2.4.3.2. Interfacial tension and surfactants

The interfacial tension, or force per unit length along the interface between the gas phase present due to bubbles and the fluid that is blood, has to be considered [244]. The Young– Laplace law relates the pressure discontinuity 훥푃 across the interface assumed static, to the mean interfacial curvature 휅, and interfacial tension 휎 [250]:

훥푃 = 2휎휅. ( 22 )

For a dynamic interface, this equation is replaced by:

훥풇 = 휎휅풏 + 훻푠흈, ( 23 )

where 훻푠 is the surface gradient operator, 풏 the normal unit vector and 풇 the traction at the boundary such that 풇 = 흈 · 풏 with 휎 the stress tensor [250]. For a Newtonian fluid, the normal stress discontinuity is balanced by the interfacial curvature term and the tangential stress discontinuity by the interfacial tension gradient. Due to boundary conditions, interface position 풀 follows

휕풀 ⋅ 풏 = 풖 ⋅ 풏 ( 24 ) 휕푥

meaning that the interface moves at the velocity of the normal component of the fluid at the interface. As the bubble starts to move and deform with the flow, the interface position becomes part of a moving boundary problem since the dynamics of the interface and fluid are coupled. The interfacial tension 휎 is typically constant for a clean interface with no temperature differences, but depends on the interfacial surfactant concentration when surfactants are present. If there is a gradient in surfactant concentration on the boundary, then this results in a shear stress gradient along the boundary, inducing Marangoni flows [244]. Surfactants or surface active species can therefore affect the dynamics of cardiovascular

90 bubbles [251]. In blood there are many lipids and proteins that are surface active which are soluble in blood and transported by convection and diffusion [252, 253].

2.2.4.4 Biological interactions

Bubbles which are formed in vivo or introduced in the organism interact with living cells. In addition to their purely mechanical implications, they induce a variety of biochemical responses. Interaction between bubbles and blood [254, 255], endothelial damage, and microcirculatory impairment [256] have all been shown.

Nevertheless studying this phenomenon precisely is difficult. Studies in vitro and ex-vivo, although insightful, are difficult to extrapolate to living cells in their complex environment. The main limitation of animal studies, on the other hand, is the higher complexity level that humans exhibit and the inability to isolate phenomena for study. There are also practical difficulties in finding adequate transparent tissue for intravital microscopy of microbubbles in blood vessels [257]. Computational and analytical studies can fill the gap between these methodologies to a certain extent; however the inflammation cascade caused by bubbles is an additional important problem to consider.

Not considering the disruption that bubbles may cause to get into the circulation in certain cases, once the bubbles have detached and are flowing in the vasculature their effect on endothelial function has to be questioned. It was found in the case of microbubbles injected in excised rat vessels for instance that their adhesion to the endothelium is significantly lowered in the presence of surfactants [258]. This could be due to surfactants acting as drag-reducing agents [259, 260] and reducing shear forces so that the mechanical stress upon the endothelial surface is reduced. Adhesion of microbubbles to endothelial cells, as well as damage to these, may trigger an inflammatory response and obstruct blood flow, causing ischemic injuries [261]. The properties of the bubble boundary layer are clearly relevant to characterize adhesion properties.

It has been shown for instance that some circulating molecules or dissolved substances in blood can adsorb on the interfacial area of the bubble altering their dynamic behaviour [262, 263]. Proteins adhering to the bubble interface can affect the adhesion interaction to endothelial cells [154, 224, 264]. In addition, they can play a role in signalling clot formation

[252]. The adhesion of gas emboli to the endothelial surface was thus shown to result from the blood-borne macromolecules attached to the bubble surface [154]. These considerations are important for the therapy of gas embolism, as it has been suggested that adding surfactants to the perfusate can reduce bubble adhesion force [265].

Nitrogen bubbles have also been observed to induce an inflammatory response. Platelets and leucocytes have been shown to aggregate with the presence of bubbles [266, 267], denature lipoproteins and activate complement, bradykinin and coagulation systems [189]. These in turn induce capillary leakiness and hemoconcentration [46, 268], which contributes to the difficulty of treating decompression sickness by means of recompression only once this inflammatory process is under way.

Ultrasound contrast agents can also be internalized by cells. For instance various contrast agent types have been shown to be phagocytosed by Kupffer cells (liver-specific macrophages) in vitro [269]. The fact that some bubbles therefore end up accumulating in the liver or spleen can be used for delayed phase imaging [190].

2.2.5 Conclusion

Bubble growth from desorption of a liquid containing dissolved gas is generally dependent on both heat and mass transfer. The case of decompression driven growth primarily dependent on mass transfer was presented for the bubbles growing endogenously in the body of scuba divers during and after their ascent from a dive. This decompression degassing was separated into the diffusion and inertia controlled bubble growth from a pool or flowing liquid respectively. The detachment equations for a bubble growing on a solid surface in a liquid were then presented with and without flow. The behaviour of bubbles in the bloodstream was considered, looking in particular at their time to dissolution under different saturation conditions and geometries, rheology and biological interactions.

Defining the exact cut-off point between the presence of bubbles in the circulation and the observation of clinical symptoms is yet unclear, in both the endogenously formed bubble case from decompression [182] or mechanical heart valves [270] and the iatrogenically introduced bubbles [153]. The persistence of bubbles in the bloodstream is crucial in these inflammatory processes and gas composition of the bubbles compared to the dissolved gases in the body

92 affects the elimination time, therefore the solubility and diffusion coefficients of these gases in the given tissues are of interest [216]. As such, physical modelling and imaging techniques that can offer estimates of numbers and sizes of bubbles, as well as shell properties information, are clearly needed. Indeed the rheological behaviour and interaction of bubbles with the different components of blood and vascular wall determine the inflammatory response triggered and ultimately the clinical presentation.

2.3 Rationale behind PhD research

2.3.1The need for hyperbaric decompression stress quantification

From the background and literature review, it is clear that although scuba diving remains a relatively safe activity with respect to DCS incidence, the risk is not uniform amongst people and the majority of recreational accidents nowadays happen despite there being no violation to current algorithms. In addition to known detrimental long term effects of professional diving [271-275], there is new evidence emerging of long term effects even for no history of DCS and even amongst recreational divers [276, 277]. Furthermore, inter-personal differences in risk are well-documented and physiological variables seem to play an important role [278].

This points to the need for developing new techniques towards the quantification of hyperbaric decompression stress. Instead of just preventing decompression sickness (DCS), the aim is to go towards an environmental cardiovascular personalised stress index, especially as sub-clinical long terms effects of even recreational scuba diving have been established. As such, a new endpoint, beyond simply DCS/no DCS, for evaluating decompression models and algorithms is needed. As there is documented evidence that a higher bubble grade post dive, assessed using ultrasound imaging, translates to higher DCS risk, one component of this new evaluation endpoint has to include post dive circulating bubbles in the bloodstream (Venous Gas Emboli or VGE). For a comprehensive measure, however, physiological markers quantifying oxidative stress [279, 280] should also be taken into account. A combination of physiological markers and VGE assessment could therefore be proposed as a revised endpoint for algorithm validation. The advantage of such an approach for researchers is that preconditioning interventions and other inter-personal physiological differences could then be assessed.

From an engineering perspective, despite the longevity of the research field, a number of fundamental issues that remain unknown have prevented efficient modelling. The aim of this thesis is to directly tackle the research methodology by developing three tailored tools. The proposed experimental and computational techniques could be used towards optimising the cardiovascular risk assessment of hyperbaric decompression stress caused by circulatory bubble dynamics.

This is of crucial importance for the effective prevention of DCS which can result from exposure to altitude, space flight and extravehicular activities, hyperbaric tunnelling works, as well as recreational, technical and commercial scuba diving. The increasingly extreme profiles undergone by individuals in this context (offshore oil excavation, recent endorsement of deeper and more technically challenging dives using different breathing mixtures, human space flight programs) calls for a systematic investigation of the underlying physics and physiology at play. Importantly, scuba diving as a model of environmental stress, incorporating pressure, immersion, temperature and exercise, is but an example of physiological adaptations on different timescales, as both short and long-term coping mechanisms are triggered during and after the dive. Having a comprehensive assessment of this stress would be particularly useful in preventing long term effects, but also looking at how effectively some of the short term effects can be reversed in healthy diving. This could in turn lead to insights for endothelial dysfunction, oxidative stress responses or emboli.

2.3.2 Defining PhD aims

In line with the above paradigm shift going towards quantification of a hyperbaric stress index, the aim for this PhD is to develop three experimental and computational tools and techniques.

Firstly, a simulation platform will be developed in MatLab to model the diving process by optimizing the implementation of current dissolved gas phase tracking decompression models. This platform will be able to simulate diving scenarios, while also analysing real dive profiles and see how real data compares to the modelling predictions, which will be investigated with data from the Divers Alert Network Europe (DAN Europe).

Chapter 3 – Development of the Modelling Framework

3.1 Aim and Scope

We have argued for the need for developing new techniques towards the quantification of hyperbaric decompression stress, redefining the goal of decompression modelling towards a personalised stress index combining physiological markers and VGE. An obvious advantage for calibrating and validating such an approach is that the outcome would not be binary anymore (DCS or no DCS) but on a continuous scale of “stress”. This in turn allows more data being collected from all dives, moving away from the limited current military (fit, young, male) square-profile databases. It also helps with devising experiments as it could be deemed unethical to use DCS-provoking profiles in volunteers for testing.

This chapter deals with the development of a computational modelling framework for current decompression algorithms implementation. The aim is to develop a research platform implementing dissolved gas phase decompression models for the purpose of comparison and testing with real data from the Divers Alert Network Europe (DAN Europe). The algorithms implemented are described, as well as the graphical user interface choices for graphical output and comparison in analysing real dive profiles.

3.2 Development of a MatLab decompression platform implementing dissolved gas phase tracking algorithms7

3.2.1 Simulation platform inputs and outputs

A MatLab8 code was written from scratch for a simulation platform calculating the decompression procedure that a diver would need to follow to ascend safely after diving at a

7 Adapted from V. Papadopoulou, Early Stage Assessment Report, Imperial College London, Bioengineering Department, 2012. 97 certain depth for a certain time. The chosen implementation strategy is that of decompression tables: the input given considers the diver spends all his time at the maximum depth and the optimal ascent scenario is calculated considering possible stops at fixed depth intervals (3m), see Figure 12. Other strategies could be considered with precise inputs (multilevel diving) but the choice here relates to the validation of the implementation, as the table method allows for direct comparison with popular (eg US Navy) tables based on dissolved gas phase tracking. A multilevel implementation is presented in the next section of this thesis when the input is the whole dive profile and the question is no longer the calculation of the ascent from depth, but rather whether the full real dive profile including ascent violated the model at any point.

Figure 12: Input and output of decompression simulations plotted on a model decompression profile.

8 All data processing was performed off-line using a commercial software package (MATLAB 7.13, The MathWorks Inc., Natick, MA, 2011) 98

A graphical user interface (GUI) to input parameters was created in MatLab, as shown on Figure 13. The rate of ascent and descent and number of compartments to use are predefined in the code.

Figure 13: MatLab GUI interface for user input (breathing mixture, bottom time, maximum depth and calculation method) and output display.

3.2.2 Basic code description

3.2.2.1 Principles

Dissolved gas phase gas tracking is based on monitoring the partial pressures and gas loading of a number of compartments defined by their halftimes which dictate the rate at which they absorb and release inert gas exponentially (the reader is referred to Chapter 1 for more background information on the decompression procedures developed by Haldane, Workman and Bühlmann).

For a given compartment at a certain time the gas loading differential equation is given by equation 25, 99

푑푃 푐 = 푘 (푃 − 푃 ), ( 25 ) 푑푡 푖푛푠푝 푐

where 푃푐 is the pressure of the inert gas inside the compartment, 푃푖푛푠푝 the inspired inert gas pressure and 푘 a time constant corresponding to that particular compartment which is found to be 푘 = 푙푛(2)/퐶 where 퐶 is the halftime of the given compartment in minutes. The 푡 analytical solution for equation 25, for any 푃 (푡), is 푃 (푡) = 푘푒−푘푡 푒푘푡′ 푃 (푡′)푑푡′. 푖푛푠푝 푐 ∫0 푖푛푠푝

A total of 16 compartments [281] were chosen, with two sets of halftimes, one for nitrogen and one for helium, see Table 2. These compartments start out at the beginning of the dive saturated for sea level breathing air (atmospheric pressure), then during the descent of the diver they absorb inert gas, until they become saturated. During ascent, they start to release nitrogen and helium back into the circulation and out of the body through the normal respiration process. Nitrogen Helium 4 1.5 8 3 12.5 4.7 18.5 7 27 10.2 38.3 14.5 54.3 20.5 77 29.1 109 41.1 146 55.1 187 70.6 239 90.2 305 115.1 390 147.2 498 187.9 635 239.6

Table 2: Table with Nitrogen and Helium half-times (in min) used for the 16 compartments, as per the Bühlmann ZH-L16 algorithm [281]

100

The tolerated partial pressure inside each compartment – where “tolerated” is assumed to mean that this will not result in DCS (or bubbles in these models, although we know this does not hold) – depends on the halftime of that compartment, but also follows a linear relationship with depth. This is the concept of M-values (threshold line of tolerated pressure) represented in Figure 14. Interestingly a certain degree of over-pressurisation is allowed for compartments also upon surfacing. This has important implications for repetitive diving, where the diver undergoes a subsequent dive before having all of his compartments return to their initial partial pressure pre-first dive. For the case of repetitive diving, the surface interval (time between dives) is therefore used to calculate updated initial partial pressures for the start of the next dive.

Figure 14: Simplified (ignoring ascent rate) graphical representation of the M-value concept, showing that the tolerated gas loading varies with depth, where point A is the start of the decompression.

The idea for the decompression procedure is thus to calculate the partial pressures of all compartments at a given time, then to calculate the “ascent ceiling” allowed for each. This is the shallowest pressure (depth) that a given compartment can be brought to without exceeding its threshold partial pressure for that depth. Then the deepest ascent ceiling of all compartments is chosen to be a “decompression stop”. Once the diver reaches that 101 decompression stop depth and the partial pressures of all the compartments are updated, he stays at that depth until he is allowed to ascend 3m shallower.

The 3m procedure is a practical way of implementing this algorithm, since smaller increments will be difficult for a diver in terms of buoyancy control in harsh conditions. The amount of time spent at a decompression stop is thus calculated by calculating the deepest ascent ceiling every minute and allowing a change of depth as soon as the new deepest ascent ceiling is 3m shallower than the last, until the surface is reached.

3.2.2.2 Individual Functions

Updating partial pressures at constant depth: For a constant depth section of a dive, equation 25 can be solved exactly since this corresponds in practice to keeping 푃푖푛푠푝 constant. This is because the breathing gas is delivered by the regulator to the diver at ambient pressure throughout the dive and the ambient pressure is constant if depth is kept constant. Therefore for each compartment the equation describing gas loading is given by equation 26.

푖 푖 −푘푡 푃푐 = 푃푐 + (푃푖푛푠푝 − 푃푐 )(1 − 푒 ) , (26)

where 푡 is time, initialised for the start of the constant depth section of the dive, 푃푐 is the pressure of the inert gas inside the compartment at time 푡, 푃푖푛푠푝 the inspired inert gas 푖 pressure, 푃푐 the compartment inert gas pressure at 푡 = 0 and 푘 the time constant corresponding to that particular compartment.

Updating partial pressures when changing depths: When the depth is not kept constant, a further assumption is used to update the inert gas pressure for every compartment: the descent rate 푅 (or ascent rate −푅) is kept constant. This allows one to assume a constant rate of 푖 푖 change for 푃푖푛푠푝 = 푃푖푛푠푝 + 푅푡, where 푃푖푛푠푝 is the initial inspired inert gas pressure. For the purpose of these simulations the chosen value is a conservative 푅 = 12 (푚푠푤/푚푖푛) in

102 accordance with the most conservative recreational scuba diving guidelines. Equation 25 can then be solved to give equation 27,

1 푅 푃 = 푃푖 + 푅 (푡 − ) − (푃푖 − 푃푖 − ) 푒−푘푡, ( 27 ) 푐 푖푛푠푝 푘 푖푛푠푝 푐 푘

푖 where 푃푖푛푠푝 is the inspired inert gas pressure at the beginning of the ascent/descent portion (for 푡 = 0). Note that setting 푅 = 0 in equation 27 gives back the constant depth equation, equation 26, as expected.

Calculating the ascent ceiling depth at a certain time: The ascent ceiling or maximum depth allowed for each compartment at a certain time t is a function of the inert gas loading of the compartment 푃푐(푡). Bühlmann introduced the coefficients 푎 and 푏 for every compartment of halftime 퐶. The M-value linear relationship of tolerated compartment inert gas pressure with depth 퐷(푚) can then be written as equation 28:

1 2 푀(퐷) = (푎 + ) + ( ) 퐷. (28) 푏 푏 where

푎 = 2 퐶−1/3, (29)

푏 = 1.005 − 퐶−1/2. (30)

The pressure ascent ceiling 푃푟푒푠푠푢푟푒퐶푒푖푙푖푛푔 at a given time 푡 and for a given compartment is then calculated using equation 31 (Figure 14),

푃푟푒푠푠푢푟푒퐶푒푖푙푖푛푔(푡) = 푏 (푃푐(푡) − 푎). (31)

The greatest pressure ceiling of all compartments is then chosen and converted into a depth ceiling. This is the depth that the diver is allowed to ascend to without violating any compartment's M-value. 103

When calculating the first decompression stop, this depth ceiling (in meters) is rounded up to the greater multiple of 3 for conservatism.

Determining time at decompression stop: This is done by updating 푃푐 using equation 26 every minute and subsequently calculating the new ascent ceiling using equation 31, until it is 3 meters shallower than the current decompression stop. This procedure is then repeated all the way up to the surface.

3.2.3 Decompression profiles

3.2.3.1 Example result

The GUI display on fig.15 shows one example of a calculated decompression profile, in this case a two decompression stop procedure at 6m and 3m. The calculations were originally done using only 5 compartments to simplify the analysis, the halftimes of which were 5, 10, 20, 40 and 75 minutes respectively.

Figure 15: Dive at 40m for 20min breathing air throughout the dive.

104

3.2.3.2 Corresponding compartment analysis

The evolution of the partial pressure of nitrogen inside every compartment (gas loading) is shown in fig.16 for the profile considered above (fig.15) and the corresponding pressure ascent ceilings calculated by the program are shown in fig.17. The compartment which limits the depth of ascent at the beginning of the decompression (when 푡 = 20 푚푖푛) is the one with halftime 퐶 = 10 푚푖푛, despite a higher partial pressure inside the compartment of halftime 퐶 = 5 푚푖푛, which is a manifestation of the different M-Values intercepts and slopes for each compartment. The threshold for which the second decompression stop (3m) is allowed is clearly visible on fig.17, where it happens as soon as the red curve goes below the pressure ceiling of 1.3 (ata, atmospheres absolute) (which is indeed 3m depth in terms of depth ceiling). Once the new depth of 3m is reached, the dominating compartment in terms of decompression is the one corresponding to 퐶 = 20 푚푖푛.

Figure 16: Partial pressure of inert gas for the 5 chosen compartments and corresponding dive depth during the decompression portion of the dive.

105

Figure 17: Corresponding ascent ceiling. At the beginning of the decompression, the diver is allowed to ascend to 6m since the highest pressure ceiling is 1.53 ata (5.3m depth, rounded up to the biggest multiple of 3, therefore 6m).

For this dive profile adding more compartments will not change anything since already with 5 compartments the 40min and 75min halftimes are superfluous. However since the dominating compartment will vary depending on the dive profile (because of the M-values slopes and intercepts), having a range of halftimes is clearly important for both very long shallow dives and deep short (bounce) dives.

3.2.4 Comparison against known tables

3.2.4.1 DSAT No-Decompression Limits

Recreational diving originally used the US Navy no decompression limits (NDL) [282]. This was however deemed to significantly limit the allowable time spent underwater unnecessarily since recreational diving is not cold, precision, physical exercise. The DSAT tables were 106 developed for use in the recreational setting to address this point and are based on the Bühlmann implementation of dissolved gas phase tracking algorithms, making a comparison useful in checking the MatLab implementation.

NDL (min) Depth MatLab (m) DSAT (PADI) platform 14 124 102 16 81 75 18 58 57 20 42 46 22 31 27 24 26 32 26 21 27 28 17 22 30 15 19 35 11 14 40 8 9 42 7 8

Table 3: No Decompression Limit (NDL) for depth exposure, calculated from the MatLab decompression platform and from the DSAT planner (PADI).

It should be noted however that although based on the Bühlmann profile they were slightly modified ad hoc based on human diving trials and expert opinion for convenience of use. This, taken together with rounding up between feet/meter conversions and whole minutes in calculations, explains the slight departure from exact values. In particular, the shallow depths times were manually lowered in the DSAT tables after human trial [282]. Nevertheless the agreement shown in Table 3 is close enough to confirm that the modelling is correctly implemented, and there is no systematic error (trend between the variation with increasing depth). The fact that these DSAT tables are the ones used in practice by the largest training agency PADI in the no-stop recreational setting makes them a very good comparison for the later comparison to real recreational diving.

107

3.2.4.2 US Navy decompression tables

For illustration purposes a few decompression dives ascents calculated using the MatLab platform and also presented here next to the official US Navy decompression tables, Table 4. It should be noted that the actual ascents are not directly comparable since the USN uses only 5 compartments (not 16 as in the Bühlmann implementation) and the ascents were changed from human trials and for practical diving reasons (please refer to Section 1.2).

The descent portion of the dive is the input for the decompression calculation in the MatLab platform, the output being the number of stops and time at each stop needed to reach the surface. Table 4 lists all USN times for 12.2m depth, then only some for the 21.4m and one for 30.5. As there are hundreds of variations this is just meant as a brief illustration of practical ascents for divers. An example of how to read the table is provided in its legend for clarity.

Descent portion of dive Decompression Procedure (staged ascent)

MatLab decompression stop(s) USN decompression stop(s) Depth (m) Bottom Time (min)

depth (m) time (min) depth (m) time (min) 200 - - - - 210 3 3 3 2 230 3 8 3 7 12.2 250 3 12 3 11 270 3 15 3 15 300 3 27 3 19 3 31 3 33 100 6 4 3 58 3 56 21.4 140 6 16 6 8 3 82 3 79 170 6 27 6 19 3 98 3 78 6 39 6 41 30.5 120 9 21 9 12 12 4

Table 4: Example decompression procedures for some square profile air dives using USN tables [50] and from the MatLab decompression platform. For example: A diver having stayed at 21.4m depth for 100min will need to make two decompression stops according to the MatLab platform (4min at 6m then 31min at 3m) and one stop according to the USN table (33min at 3m).

108

3.3 Analysis of real dive profiles (dissolved gas phase)9

Section 3.2 deals with the development and validation of a MatLab decompression modelling platform implementing dissolved gas phase tracking algorithms to calculate the ascent of a diver from depth. In so doing, functions for updating the partial pressures of compartments along dive profile segments either at fixed depth or for varying depth but with fixed speed of ascent/descent were created. These can be used directly to calculate the partial pressures of compartments during a whole dive, using now as input to the platform the whole dive profile (not only the descent section) to look for potential violations of the dissolved gas phase tracking decompression modelling principles. In effect, this allows analysing databases of dives tracked by dive computers (where we therefore have access to the full dive profile). If these databases contain DCS cases, it is then possible to ask whether significant differences in the degree of conservatism of the corresponding dive profiles took place. In the absence of any other physiological information, the degree of conservatism here is taken simply as the degree of coming close to, or surpassing, the limits set by the dissolved gas phase tracking algorithms. It should be noted that profiles which do not result in DCS are also noteworthy of analysis as they can give an insight into whether coming close to or violating the guidelines results systematically in DCS. The important but difficult additional consideration is to deal with an explicit quantification for this “degree”: should we consider the single worst partial pressure of any compartment at one particular time, or take into account how many instances during the dive profile some compartments were close to the limit, account for the temperature of the water, etc?

With the above considerations in mind, the aims here are to first modify the MatLab program (and validate this modification) to read dive profiles automatically to calculate and store instant partial pressures of inert gas for all compartments throughout the dive. The second aim is then specific to the section of the database we have access to: propose a preliminary analysis of a section of the Divers Alert Network (DAN) recreational diving database of dive profiles which includes some accidents.

9 The analysis of real dive profiles was the subject of an MSc project by Chris Song which I supervised directly. Please refer to the acknowledgements section. 109

3.3.1 Methods

The first step in analysing real dive profiles is inputting the data into MatLab from the DAN database. The platform is also modified to track partial pressure of compartments from start to finish of the dive, taking care to account for repetitive diving, then different conservatism parameters calculated. Finally, these are used to compare two different groups from the database: one resulting in diagnosed DCS and one of confirmed uneventful dives.

3.3.1.1 Harmonization of data format entry

The Divers Alert Network (DAN) Europe database started collecting recreational dive profiles since 1998 and is the largest of its kind [21]. There are in total over 35000 profiles, recently organized in a comprehensive Microsoft Access database linking all the information corresponding to each dive with other data in different sheets using a unique dive ID that includes in its name the date and time of the start of the dive.

We have access to 105 dive profiles in total, as well as all the 35 000 other sheets of information. The 105 profiles were specifically chosen so that 20 of those are confirmed DCS cases, and 85 were uneventful dives, all randomly chosen so that the years spanned and demographics did not differ significantly between the two groups.

The profiles will be analyzed with the decompression modeling platform implemented in MatLab, but first some additional demographical data is also extracted from the database for reference: average maximum depth and time of the dive, average age and BMI of divers, percentage female divers, number of dives below 40m (extended range).

3.3.1.2 Modifications and validation for analysis of real dive profiles

The dive profiles in the database come from the logged info on the divers’ dive computers, which has been harmonized so that the data all appear as successive excel entries in the DAN Access database, each linking to a unique dive ID. A computer logs data every 20s so successive data points are 20s apart and this is used to calculate the “real-time” ascent/descent rate of the diver.

110

In the case of non-repetitive diving all compartments are initialized for 0.79 pp of Nitrogen and 0pp of Helium, otherwise these initial values are calculated according to the previous dives and time at surface between dives.

The differential equations solutions for constant depth or constant speed of ascent/descent are then used to update the partial pressures in all compartments throughout the dive. These are stored in an output file for each timepoint for further analysis.

In order to validate that the MatLab program correctly tracks the values of the partial pressures in compartments during a dive, a simple square descent profile of 40min to 40m with air is used, where the ascent was calculated using the MatLab platform. The partial pressures tracked on the full profile should match the predicted ones exactly and each depth change should correspond to an ascent ceiling as calculated. Figure 18 shows that the tracking of the partial pressures of the compartments match the ones predicted when calculating the ascent using the dissolved gas phase tracking algorithm.

Figure 18: Nitrogen partial pressure for the compartments for a theoretical profile of 40min to 40m with air calculated using the decompression platform in Section 3.2 from bottom time and depth input (least conservative ascent, in black). In colour: calculated with the analysis platform throughout the full dive

111

For further analysis, we also calculate the instantaneous “degree of conservatism” for the compartments partial pressures at each timepoint. This is taken as the ratio of the pressure ceiling over the ambient pressure (same thing as the theoretical maximum tolerated gas pressure over the partial pressure of inert gas for that depth). By definition, the degree of conservatism is therefore at each time point: below 1 for safer dives, exactly at 1 for the limit, or above 1 if violations occur.

Similarly for validation, looking at the conservatism of the above dive profile (40min to 40m, air theoretical least conservative profile) should show values very close to the limit allowed since it is a profile calculated for fastest ascent without violating the limits. Setting the degree of conservatism as 1 for the case where it matches but does not exceed the limit, all values of the leading compartment (so worst degree of conservatism) should always be close to 1, as confirmed by Figure 19. The reason why it is not exactly one is numerical: since the decompression platform updates every minute. The worst partial pressure ratio is always close to 1 once the ascent begins and for every depth change since these are calculated for the least conservative (but permitted) ascent (in fact worst value is 0.9995 due to rounding).

112

Figure 19: Top: Depth (red) and worst conservatism ratio (blue) throughout the dive; Bottom: percentage of breathing gases throughout the dive, useful in case of technical diving with gas switching.

3.3.1.3 Statistical analyses

The MatLab program for analysing real dive profiles with respect to dissolved gas phase tracking is used on the 20 accidents and 85 uneventful dives, taking care to account for the cases of repetitive diving in the initial values of partial pressures for compartments.

After normality test (Kolmogorov-Smirnov (K-S) test, D’Agostino’s K-squared test and Shapiro-Wilk test), the values of the variables of interest are compared between the two groups with either a Student T-test or Mann-Whitney U-test (non-parametric). Statistical significance is set at p<0.05.

113

The variables of interest are the following:

 Average ascent rate and fastest instantaneous ascent rate  Average and worst degree of conservatism and corresponding leading compartments  Percentage of the ascent (time) spent at less than 20% conservatism  Percentage of the ascent (time) spent at less than 0% conservatism

3.3.2 Results

3.3.2.1 Epidemiological findings

Some broad statistics were extracted from the full DAN Europe database (without access to the individual dive profiles) in order to document the sort of dives and divers that are encompassed in this database.

On calculating the age of the divers, some obvious mistakes were spotted (age, calculated from the data of the dive and date of birth of the diver, of 157 or -1 for example) so these profiles (433 in total) were ignored in the following statistics. There are also missing values in some of the parameters for each dive recorded, resulting in a different total number of cases for each parameter considered below. The data is presented as means ± standard deviation.

Total number of different divers: 13 771

 Average age: 38.5 ± 9.3 years  Female divers: 2106 (14.8%)  BMI of divers (only available for 8548 divers from height and weight): 25.5 ± 10.2

Total number of dive profiles: 35 700

 Average depth: 28.2 ± 13.6 m  Average dive time: 44.1 ± 43.2 min  Number of dives below 40m depth: 5883 (16.5%)

114

3.3.2.2 Relationship between accidents and degree of conservatism

Figure 20 and Figure 21 show the statistical comparison between the no DCS and the DCS groups for worst degree of conservatism during the dive and fastest ascent rate respectively. No significant difference is found for the degree of conservatism (Student T-test, p>0.05) and it is interesting to note that even for the DCS group no violation to the dissolved gas phase algorithm was found (conservatism degree above one). However, there is a significant difference between the fastest ascent rates of the two groups (Mann-Whitney U-test, p<0.05).

Figure 20: Comparison of worst degree of conservatism (Student T-test) between DCS (1.28 ± 0.27) and no DCS (1.23 ± 0.14) groups; no significant statistical difference (p=0.456).

Figure 21: Comparison of fastest ascent rate (Mann-Whitney U-test) between DCS (12.0 ± 4.4 m/min) and no DCS (9.1 ± 3.7) groups; significant difference (p=0.0047)

115

3.3.3 Discussion

The dissolved gas phase tracking decompression algorithm implementation presented in Section 3.2 was modified so that real dive profiles can be analysed with the platform, updating the partial pressures of dissolved gasses in all compartments throughout the dive and storing these for further analysis. The degree of conservatism of the dive, defined as the ratio between real and theoretical least conservative permitted ascent for each portion of the dive, was proposed as a very simple measurement endpoint for probing real data.

From a first analysis of real dives from the DAN Europe Database, comprising of 85 uneventful dives and 20 confirmed DCS cases, no difference in the degree of conservatism of the dives was found. In fact, none of the accidents had violated the dissolved gas phase hard limits of ascent ceilings at all. This is particularly interesting, although not surprising as well- reported in the literature, since the dissolved gas phase implementation is the most common in dive computers (specifically the Bühlmann ZHL16 used here). Although artificially “tweaked” to add some conservatism for altitude, cold, etc, these models only track a calculated partial pressure of dissolved gas in theoretical compartments and thus do not account for any physiological differences between subjects.

We lack physiological data from the divers whose dives were analysed in this study, so inter- personal differences would be speculative. This is the biggest limitation of this analysis, as VGE post-dive or FMD, etc are not available. The profiles shared with us had no linked diver information, but they were randomly selected then manually checked for roughly similar divers (all male, matched age groups) and the matching of the two groups was done so that: all dives analysed are air dives and 10/20 were repetitive dives in the DCS group versus 54/85 in the non-eventful group. The results however clearly show that accidents do happen without any violation of the algorithms, and therefore in this range other factors are at play. Previous literature has for instance correlated bubbles post dive to higher BMI and older age [283] and gender differences have also been demonstrated [284].

Although no link was found in this dataset between DCS and dissolved gas phase model, there was a significant difference in the ascent rate between the two groups (p<0.05, Mann- Whitney U-test): DCS group 12.0 ± 4.4 m/min and no DCS (9.1 ± 3.7). If the results of this small sample are representative, this could be an argument for lowering the permitted ascent rate. The question of allowable ascent rate has been discussed for decades [285-289], and the

116 biggest recreational diving training agencies differ in their stated highest permitted ascent rates. Recommended guidelines from training agencies quote maximum ascent speed between 9msw/min and 18msw/min: PADI, the Professional Association of Diving Instructors, lists 18 msw/min; BSAC, the British Sub-Aquatic Club, 12msw/min); and the USN (United States Navy) and NOAA (National Oceanic and Atmospheric Association) 9.1msw/min (30ft/min).

Experiments on 50 male recreational divers showed that a 17m/min ascent yielded a higher bubble grade post dive compared to a 9m/min ascent [290]. Using DCS severity as an outcome, studies on rats also showed that ascent rate was a significant factor [291], however more recent studies with DCI as an endpoint were less conclusive due to confounding factors [292]. Anecdotally, increasingly technical divers also refer to the “optimum” ascent speed around 9-10 msw/min (quoting from personal experience “less fatigue” post dive) [293], and claim that an ascent could also be “too slow”. This is often a misinterpretation of a study looking at bubbles post dive for a 3m/min and two 9m/min ascents: the 3m/min ascent did indeed find more bubbles, but the comparison is not fair since both 9m/min rates were done with safety stops for a total time of 5 or 10min. Nevertheless, the question of assessing the optimum ascent rate remains somewhat open.

3.3.4 Conclusion

In conclusion Chapter 3 presents a MatLab decompression platform for calculating the required decompression, or analysing real dive profiles, with respect to dissolved gas phase tracking decompression models. An analysis of real dive profiles from the DAN Europe Database confirmed previous reports that accidents happen even with no violations of the decompression algorithms and reinforce the need for physiological data collection and new physiological measurement endpoints to inform the modelling. In the range of data analysed the worst degree of conservatism does not play a significant role in causing DCS, however the ascent rate might. Suggestions for additional safety from our findings would be to lower the permitted ascent rate to 10 msw/min if further analysis with more profiles confirms the results. In this respect, a study looking at the effect of ascent rate of matched divers and using an array of physiological markers would be most useful (VGE, etc).

117

With the above considerations in mind, Chapter 4 and Chapter 5 deal with the development of new methodologies for bubble growth study and post-dive assessment.

118

Chapter 4 – Development of Novel Experimental Set-up 10

Vascular gas bubbles are routinely observed after scuba dives using ultrasound imaging, however the precise formation mechanism and site of these bubbles are still debated and growth from decompression in vivo has not been extensively studied, due in part to imaging difficulties. An experimental set-up was developed for optical recording of bubble growth and density on tissue surface area during hyperbaric decompression. Muscle and fat tissues (rabbits, ex vivo) were covered with nitrogen saturated distilled water and decompression experiments performed, from 3 to 0 bar, at a rate of 1 bar/min. Pictures were automatically acquired every 5 s from the start of the decompression for 1 h with a resolution of 1.75 μm. A custom MatLab analysis code implementing a circular Hough transform was written and shown to be able to track bubble growth sequences including bubble center, radius, contact line and contact angles over time. Bubble density, nucleation threshold and detachment size, as well as coalescence behavior, were shown significantly different for muscle and fat tissues surfaces, whereas growth rates after a critical size were governed by diffusion as expected. Heterogeneous nucleation was observed from preferential sites on the tissue substrate, where the bubbles grow, detach and new bubbles form in turn. No new nucleation sites were observed after the first 10 min post decompression start so bubble density did not vary after this point in the experiment. In addition, a competition for dissolved gas between adjacent multiple bubbles was demonstrated in increased delay times as well as slower growth rates for non-isolated bubbles.

4.1 Aim and Scope

Previous experiments on decompressed rats [294-297] looked at bubble growth/shrinkage on muscle and fat depending on the gas breathed. However, the main difficulty to date has been that the set-ups do not allow for real time observation of bubble growth during the decompression and tissues/animals have to be taken out of the chamber to be observed. For

10 Adapted from V. Papadopoulou et al. Colloids Surf B Biointerfaces. 2015; 129:121-129. 119 animals, the difficulty in locating bubbles means that bubbles often have to be injected [297]. For tissues, in addition to the potential problem from dislodgment of bubbles with movement (opening chamber and taking samples out), the observation is then often done in non-ideal conditions such as waiting for the bubbles to float [105, 214, 298].

The first aim of this study is thus to develop a new experimental set-up to allow for the first time real time observation during decompression of bubble growth from desorption of inert gases out of solutions on tissue substrates. In order to quantify the different parameters that could influence inception delay times, growth rate, detachment and multiple bubble behaviour, this set-up should allow for different gases and liquid compositions, temperature control, as well as optical recording of both bubble density per unit surface and precise bubble growth rate.

Higher subject fat percentage has been demonstrated as a risk factor for higher bubble grades post dive [290, 299]. Although not fully understood, adiposity as a risk factor for developing decompression sickness (DCS) has been discussed for decades. It is usually attributed to either nitrogen solubility and diffusion arguments, or fitness-related variability in subjects’ ability to cope with bubbles. We hypothesize that, in addition to nitrogen uptake and subject cardiovascular fitness, the hydrophobicity of adipose tissues may facilitate bubble growth during decompression. The second aim of this study is therefore to investigate the role of the tissue substrate (surface), mainly of fat and muscle ex vivo rabbit tissues since fat and muscle have been used most in newer decompression modelling efforts [126, 127, 300, 301], in the decompression induced bubble growth rate and density with nitrogen taken as the inert gas.

4.2 Development of experimental set-up and analysis code

4.2.1 Schematic of full experimental set-up developed

Figure 22 and Figure 23 show the different components of the set-up which allows for the first time the observation in real time during decompression of the growth rate of selected bubbles, but also bubble density per unit surface area, comparing how these vary for different tissue surfaces, decompression profiles, gas saturated liquid composition and temperature. Two configurations for optical acquisition were optimized to allow for subsequent semi-

120 automated data analysis using image processing techniques: for growth data of individual bubbles maximum magnification and lighting straight from the back of the chamber, and for density data with lesser magnification and acquisition at an angle to look at a number of bubbles per surface area (calibrated with millimetre paper to get field of view precisely). The resolution of the optical system was assessed from calibration against a known thickness wire of 56 μm.

Figure 22: Photograph showing the optical acquisition system and the temperature controlled small pressure chamber, as well as an example result.

121

Figure 23: Top view schematic of experimental set-up, showing liquid (in blue) and gas (in red) pressure flow systems.

122

4.2.2 Development of novel experimental set-up

4.2.2.1 Chamber modifications and testing

A pressure chamber was modified so that pressurisation could be controlled with the adjustable pressurised air supply in the lab, then systematically checked for leaks on pressurisation. The faulty joints and taps were replaced, o-ring resealed and the pressure and temperature readings calibrated and verified against diving measurement equipment. The rate of decompression was also adjusted for the equivalent of 10 msw/min to be used in the experiments.

Figure 24 : Back view of the chamber before subsequent changes (but already fitted with new airtight connectors)

4.2.2.2 Saturated liquid and inlet into/ outlet out of chamber

The saturation of the liquid at the desired pressure was achieved by connecting a liquid tank with the pressurised gas to saturate it with. The monitoring of its pressure allowed us to disconnect the inlet valve once this did not trigger a fast decrease (achieved saturation). 123

Figure 25 : Fitted connectors for the tube (red here) to allow the liquid flow in and out

Figure 26 : Front view of the chamber showing the tube that lets the saturated liquid in the chamber

4.2.2.3 Temperature control: sensors, heating belts and PID

Two thermal sensors were fitted to the experimental set-up: one in the pressure chamber to be in direct contact with the tissue surface and one in contact with the gas saturated liquid in the tank. Heating belts were fitted around the pressure chamber and the tank where the liquid saturates. In order to allow temperature regulation and control throughout the experiment, both sets of heating belts are connected to proportional–integral–derivative (PID) devices 124 with set temperatures to reach that control the state (on or off) or the plugs in which the belts are connected. The temperature regulation was shown possible ± 1ºC even during decompression.

Figure 27 : Liquid saturation tank fitted with heating belt for temperature control

Figure 28 : Front (left) and back (right) views of the chamber once fitted with the purpose-built heating rings. All ring are plugged into the same multi-plug switched on and off by a PID device (temperature control).

125

Figure 29 : Temperature control overview of chamber set-up

Figure 30 : Thermal sensor fitted to the chamber (airtight purpose-built seal): inside front view (left) and outside back chamber view (right)

126

4.2.2.4 Design of inside container and emptying mechanism

To allow the observation of bubbles growing from decompression on tissues, a container was designed and built to provide support and adjustable positioning, so that the optical acquisition during the decompression could be achieved. The design takes into account the curvature of the chamber floor and allowed for its height and tilting to be adjustable (Figure 31). A grating inside the chamber allows for tissue support whilst not preventing the smooth filling of the container with gas saturated liquid. The hole in the base allows for the liquid to be drained without opening the chamber from the back tap, and the temperature sensor can be clipped on one side of the glass to stay in position throughout the experiment.

Figure 31 : Container and base, front view as it goes in the pressure chamber (left) and back view (right) showing the water inlet for the saturated liquid; also used to drain

127

Figure 32 : Emptying container system

4.2.2.5 Optical data acquisition set up and lighting

After numerous lighting positioning trials, providing light from the back of the chamber with a diving torch and adjusting the intensity, distance and angle was found to give the best contrast around the bubbles (crucial for automating the analysis of bubble growth thereafter by finding the boundary, impossible if tissue is vertical and lighting from the front).

Figure 33 : lighting set-up (chamber view from the front, open glass door removed)

128

Figure 34 : Optical acquisition set-up for bubble growth data (front view)

4.2.3 Experimental procedure

4.2.3.1 Detailed Steps

The experimental steps for each experiment are as follows:

 Saturate distilled water with nitrogen after liquid has reached and is maintaining temperature (connecting liquid tank to pressurised gas tank with mid-pressure of 3bar)  Set-up tissue substrate in container and support, including water inlet and thermistor, as well as camera lighting from the back + thermo-regulation  Close pressure chamber and set-up optical acquisition system from front checking focus distance etc (and allowing for water change) 129

 Pressurise chamber after the pressurised inlet from next room is regulated to just above 3bar and bring chamber to 3bar then maintain  Open valve for liquid to flood inside tissue container (liquid is at 3.1bar and chamber at 3bar), slowly  Once liquid reaches level wanted in container close inlet.  Refocus optics for acquisition (cf water refraction) and start timer for automatic acquisition every 5 seconds to coincide with start of decompression  Do decompression profile as per experimental plan  Keep taking pictures for 1 hour post decompression start (either growth or density focus as per experimental run) and check temperature does not display changes.  After the end of the experiment, recompress the chamber a little bit (0.3 bar) and then turn valve for liquid outlet (the pressure difference will empty the chamber avoiding flooding which is impractical to clean with the electric circuits around and then fogs up the glass for any subsequent experiments), depressurise, open chamber and clean everything and let dry.

4.2.3.2 Camera acquisition configuration for bubble growth and density

Two configurations were chosen: for growth data of individual bubbles maximum magnification and lighting straight from the back, and for density data with lesser magnification and acquisition at an angle to look at a number of bubbles per surface area (calibrated with millimetre paper to get field of view precisely). In addition, the field of view obtained with the original camera set-up did not permit a good enough resolution to see small bubbles being formed so new extension rings were added.

130

Figure 35 : Automated acquisition control for picture acquisition every 5 seconds

Figure 36 : Picture of a trial configuration for resolution testing

131

Figure 37 : Camera set-up for density acquisition

4.2.4 Image processing analysis code in MatLab

A custom MatLab image processing code was written for bubble tracking in successive frames, based upon a Hough transform for circle pattern recognition [302-304]. The program identifies the bubble radii and center positions, tracking those over a time sequence of pictures, then outputs the results in an excel file for analysis. In addition, figures displaying the original picture and overlaying the bubbles centers and radii are also created and saved. The program successfully tracks bubble growth over time. Conflict resolution if multiple radii and/or centers are found, as well as consistency checks, are implemented to ensure correct results are output to the excel file. The main steps of the code are shown on Figure 38.

The sequence for bubble recognition using the Circular Hough Transform feature extraction is fully automated. However, in practice it is convenient to have the user select a small range of radii as the algorithm input in order to accelerate the running time of the program which can take tens of minutes when the sequence of picture is long for a search over a wide radius range (over 50 pixels). A simple GUI was therefore created to speed up the process: it asks the user to manually select all the pictures to analyze for a sequence (with possibility to input more than one sequence at a time), displays the first and last pictures for the user to draw the bubble outline and input the time delay between successive pictures (taken 5s apart in our experiment), and initialize the algorithm from these user inputs (namely region of interest and radius range).

132

Figure 38 : Flowchart of analysis code in MatLab. CHT: circular Hough transform; DMP: decision making process to keep only one radius and centre per bubble if multiple choices given in CHT output, by taking into account the previous frame output for centres and/or ROI and/or radius value (depending on user choice for DMP input parameters).

133

4.2.4.1 Circular Hough Transform for bubble recognition (CHT)

A program was written in MatLab to automatize the analysis based on the image processing technique of the Circular Hough Transform for circle recognition (bubbles here). The Circular Hough transform method for circle detection is based on the gradient field of the image. The program finds the bubble radii and centre positions, tracking those over a time sequence of pictures, then outputs the results in an excel file for analysis. In addition, figures displaying the original picture and overlaying the bubbles centres and radii are also created and saved.

Figure 39 : Example output from MatLab program, showing the result after circular Hough-transform (right, axes in pixels) and the output with the centre and radius of the bubble marked on the original image (left, axes in pixels).

134

Figure 40 : Example of bubble recognition with bubble outline drawn in blue and centre position displayed in red (axes in pixels)

4.2.4.2 Robustness additions (DMP)

The recognition part of the algorithm is based on the circular Hough-Transform as explained above. The program successfully tracks bubble growth over time. Conflict resolution if multiple radii and/or centres are found, as well as consistency checks are implemented to ensure correct results are output to the excel file, and occasionally if in doubt the problematic picture is not closed automatically to allow the user to check it manually as well before closing.

4.2.4.3 GUI implementation for semi-automatisation

The main sequence for bubble recognition using the Circular Hough Transform feature extraction is fully automatized. However, in practice it is convenient to have the user select a small range of radii as the algorithm input in order to accelerate the running time of the program which can take tens of minutes when the sequence of picture is long for a wide radius range.

In addition, as the experimental set-up developed will be used by other PhD students in the future, who do not-necessarily have coding experience, a simple GUI was created to ask the user to manually select all pictures to analyse for a sequence (with possibility to input more than one sequence at a time), display the first and last pictures for the user to draw the bubble

135 outline and input the time delay between successive pictures (normally 5s for my experiments), and initialise the algorithm from these user inputs.

4.2.5 System Evaluation

The experimental set-up allows the accurate and sensitive optical recording of bubble growth and density on tissue surface area measurement (field of view of 6.38 mm × 4.25 mm, with a resolution of 1.75 μm) during hyperbaric decompression. The semi-automated analysis code is able to track bubble growth sequences successfully (with a maximum associated radius error of ±2 μm) and implementation time is significantly reduced from the user initialization of the radii range.

136

4.3 Experiment and analysis fat/muscle

4.3.1 Material and methods

4.3.1.1 Experimental Procedure

After Ethics Committee approval (59001/536) in accordance to EU Directive 2010/63/EU for animal experiments, ex-vivo muscle and fat tissues from rabbits were obtained for the experiments and all animal handling was done by our collaborators in the veterinary school (see acknowledgements). Rabbits (male, 3 months old, n=8) were anaesthetized then killed by injection of sodium pentothal and potassium chloride respectively. Excised tissues were then extracted within 15 minutes and placed directly in Ringer’s solution. These were stored at 4°C for preservation and used in the experiments within 48h. The tissues were covered with nitrogen saturated distilled water and eight decompression experiments performed, from 3 bar to 0 bar, at a decompression rate equivalent to 10msw/min as described below.

The following description relates to our particular experiment, however different gases, liquids, tissues and temperatures could be used with this set-up. A compressed nitrogen tank is used to saturate distilled water at just above 3 bar (taking reference of 0 bar atmospheric pressure) in the water liquid saturation tank which is temperature controlled with a PID device at 25°C. The tissue is carefully placed in the glass container in the chamber, after switching on the dive torch used for back lighting the tissue and bubbles in order to get good quality images for image processing thereafter (black bubbles on a light background). The pressure chamber, also temperature controlled using another PID device at 25°C, is pressurized to 3bar above atmospheric pressure using compressed air. The liquid saturation tank valve is then opened for the liquid to flow into the pressure chamber through a tube connected to the custom made glass compartment with the ex-vivo tissue. The liquid entry valve of the chamber is then closed again once the tissue is covered completely. The camera used for optical acquisition is refocused (water entry changes the refraction) and automatic acquisition every 5 s started to coincide with the start of the decompression, done at an equivalent rate of 10 msw/min without interruptions from the initial saturation of 3 bar (40m depth equivalent) to 0 bar (surface pressure) with a needle valve.

4.3.1.2 Theoretical analysis for bubble growth 137

Gas bubble growth on a substrate due to decompression is a complex physicochemical process. A detailed mathematical formulation is not given here but the key issues of the process are described. The difference between the instantaneous gas concentration in the liquid phase from the gas solubility (equilibrium, i.e. maximum, dissolved gas concentration) at the specific thermodynamic conditions (temperature, pressure) is the driving force for bubble growth. The motion of the gas–liquid interface as a bubble grows induces motion to the surrounding liquid. Gas has to be transferred inside the liquid flow field by diffusion and convection in order to reach the bubble surface. The pressure in the bubble increases with respect to the liquid pressure by the Laplace pressure term (surface tension effect) and the pressure needed to support liquid motion (as described by the Rayleigh–Plesset equation [305]). The shape of the bubble is spherical for diameters less than 1 mm [305] (more specifically spherical segments for non-zero contact angle). Finally, the bubble interior consists of desorbed gas and vapour (to account for vapour pressure). In addition to the above general considerations, a total balance of the dissolved gas is needed for the particular experimental set up used here since gas is consumed gradually due to creation and growth of bubbles.

In principle, a bubble growth mathematical model must account for all the above issues. Fortunately a great deal of simplification is possible under the present experimental conditions: i) Τhe experiments showed bubble growth rates of the order of few micrometers per second. For such growth rates the overpressure needed to generate the liquid motion is insignificant and the bubble pressure is simply the sum of the ambient and the Laplace pressure [115]. ii) The water vapour pressure at the experiment temperature of 25°C is less than 0.03 bar which means that the water vapour molar fraction in the bubble is less that 3% so it can be safely ignored. The bubbles can be assumed to consist only of gas. iii) The measured contact angles are relatively small so the bubbles can be assumed to be approximately whole spheres (and not spherical segments). iv) The relative significance of convection and diffusion of the dissolved gas on bubble growth is determined through the so called Foaming number [116]. The Foaming number is given as (퐶 − 퐶푒푞)/휌푔 where 퐶 is the initial dissolved gas concentration (at 4 atm), 퐶푒푞 the equilibrium dissolved gas concentration (gas solubility) at the final pressure (1 atm) and 휌푔 138 the gas concentration in gas phase at the final pressure. These assumptions are possible due to the fact that the decompression time is small compared to the experiment time and it can be assumed that the bubble growths occur after the decompression is over, under a constant pressure of 1 atm. v) A global balance of the dissolved gas considering the nucleation rate, the bubble size and the dissolved gas concentration reveals that the reduction of the dissolved gas concentration during the experiment is small and can be neglected (due to large liquid volume so 퐶 remains constant during the experiment).

The Foaming number in the present experiment is therefore found to be 0.045. This small value indicates domination of diffusion over convection which means that the latter can be ignored. Discard of convection is a major simplification which disconnects the liquid flow field from the bubble growth problem. The above assumptions may lead to an error of a few percent which is below the resolution capabilities of the experimental technique so there is no need to relax them by employing far more complex models.

Solution of the quasi-steady diffusion equation in the liquid domain combined to a bubble gas balance leads to:

4훾 푑푅 훼퐷(퐶−퐶푒푞) (휌푔 + ) = , ( 32 ) 3푅푅푔푇 푑푡 푅

where 푅 is the bubble radius, 푅푔 is the gas constant, 푡 is time, 푇 is temperature, 퐷 is gas in liquid diffusivity and 훾 is the surface tension. A mathematical analysis can show that the effect of surface tension term for bubbles with radius larger than 20 μm is comparable to the resolution of the experimental data so it can be safely ignored leading to:

2훼퐷(퐶−퐶 )푡) 푅 = √ 푒푞 , (33) 휌푔 The parameter 훼 is of particular importance and it is related to the geometry external to the bubble domain (i.e. it results from the solution of the diffusion equation in this domain). Two exact values are 훼 = 푙푛(2) [306] for bubble growing on a flat substrate and 훼 = 1 for bubble growing in infinite liquid domain. In principle, the relative curvature of the substrate

139 can change as bubble grows but this variation is general small so a constant in time 훼 has been assumed. According to the above analysis, the only reason for observing differences in the growth rate is the local geometry of the substrate around the nucleation site. There is an exception to the above statement in case of proximity of two nucleation sites (two adjacent bubbles). The concentration field around the bubble diminishes as 푅/푟 (푟 is the radial distance from the bubble) so two bubbles at short distance from each other exhibit reduced growth rates.

4.3.1.3 Data Fitting and Statistical Methods

According to the theoretical analysis the bubble radius evolution curve has the form 푅 = 퐴푡0.5. Direct use of the above equation to fit experimental data requires the exact knowledge of the bubble creation (inception) time. This is not possible due to the singular character of the growth equation at 푡 = 0 (infinite growth rate) and the experimental finite bubble detection ability. So it is found very fruitful to estimate the bubble creation moment by using it as fitting parameter.

Bubble radii over time were fitted with a constrained power law as it is expected [3] that the mass diffusion growth curves should follow the form:

푅 = 퐴푡0.5, (34)

where 푅 is the bubble radius, t time taken from the inception of the bubble and 퐴 is a constant.

The exact inception (time-of-onset of nucleation) time 푡 = 0 is not known since there is a resolution limit to the optical set-up, as well as a time delay of 5 s between successive pictures. This inception time, 푡0, is therefore extracted from the observed time-dependence of the bubble radius, by re-writing equation 34 as:

0.5 푅(푡) = 퐴(푡 − 푡0) . ( 35 )

140

In order to fit growth curves to equations 34 and 35 and estimate the 퐴 coefficients as well as the goodness of fit (퐺), the exact bubble inception time 푡0 is therefore estimated by squaring both sides of Eq. (3b) and fitting a linear equation with constrain 훽 > 0,

푦 = 훼푡 + 훽 ( 36 )

−훽 where by design 푦(푡) ≡ 푅2(푡), 퐴 ≡ √훼 and 푡 ≡ . All data is presented as mean ± 0 훼 standard deviation. Statistical comparison tests between the muscle and fat quantities measured were performed with Mann–Whitney U test after negative normality test. Statistical significance levels were set at p < 0.05 (*), p < 0.01(**) and p < 0.001 (***).

4.3.2 Results

An experimental set-up was developed to allow for the observation of bubble growth rate during decompression on ex vivo tissue surfaces. A total of 22 experiments were performed without any decompression stop (3 bar to 0 bar at a rate of 1 bar/min): 11 with fat tissue substrate (6 with the camera focused for growth and 5 primarily with the camera focused on density) 11 with muscle tissue substrate (6 with the camera focused for growth and 5 primarily with the camera focused on density). The bubble growth observed from the experiments happen in a cyclic manner from nucleation sites. The phenomenon observed is typical of heterogeneous nucleation where a bubble starts growing from a preferential site (nucleation site) until it finally detaches and floats, then another bubble grows from that same site, etc. in a cyclic manner [307].With the 12 experiments where the camera was focused for growth sequences, a total of 74 bubble growth sequences were observed (35 on fat tissue substrate of which 20 contained multiple bubbles in the field of view, and 39 on muscle tissue substrate of which 25 contained multiple bubbles in the field of view). There were also 16 sequences of bubbles growing from below camera resolution, 15 sequences observed until bubble detachment and a total of 23 sequences with good enough quality on the substrate to extract contact lines and contact angles.

141

4.3.2.1 Bubble Density

A significant difference in density of bubbles formed on muscle and fat was found, respectively 4.2 ± 1.5 and 8.4 ± 3.4 cm−2 (Figure 41), with significance set at p < 0.05. The densities were measured at 30 min post decompression start since the cyclic growth observed means no new nucleation sites are observed in practice after about 15 min post decompression start, and the bubbles per surface area stay roughly constant throughout the experiment after that point (minus bubble floating before new bubble appears for some seconds). It should be noted that the nucleation sites are not equally distributed on the tissue surfaces and some areas have more densely packed bubbles than others.

Figure 41 : Bubble density comparison between fat (n = 11) and muscle (n = 11) tissue substrates

4.3.2.2 Growth Rate

All bubbles observed growing from below resolution were fitted to a constrained power law, as it is expected [4] that the mass diffusion growth curves should follow the form: 푅(푡) = 퐴 푡0.5, where R is the bubble radius, t time taken from the inception of the bubble and 퐴 is a constant. Goodness of fit was very good (G coefficients for muscle and fat were 0.95 ± 0.076 and 0.97 ± 0.036 respectively), and no significant difference between the two tissues was found for the A coefficients (9.8 ± 4.1 and 9.0 ± 1.2 respectively for muscle and fat tissue substrates).Taking all single bubbles observed, not only the ones observed from

142 below resolution size, fitting with the radii with the expected mass diffusion growth curves showed a goodness of fit such that G coefficients for muscle and fat were 0.98 ± 0.02 and 0.97 ± 0.03 respectively, and again no significant difference between the two tissues was found for the A coefficients (11.0 ± 8.57 and 12.4 ± 8.73 respectively for muscle and fat tissue substrates). The super-imposed bubble growth curves for both muscle and fat tissue substrates from calculated bubble inception time are shown in Figure 42. Using the experimental conditions and the physical parameters of nitrogen, the value of A for the present experiments is found to be 12.6 훼0,5 μm/푠0,5. Considering the typical geometry dependent range of 훼 (0.7–1), it is clear that this value is very close to the average experimental values. The theory predicts the magnitude of 퐴 and the small difference between its value for different substrate geometries. On the other hand, the experimental scatter in 퐴 values cannot be explained at present. Arguments regarding the local consumption of dissolved gas (in contrast to the global consumption which has been deemed insignificant) seem plausible to justify a range of A values instead of a single value. Interestingly, modifications of the nucleating sites by geometric abrasions on a controlled surface were shown to affect bubble detachment size as well as produce deviations from Scriven behaviour [308].

Figure 42 : Superimposed muscle (red) and fat (blue) radius (μm) versus time (s).

143

4.3.2.3 Detachment Size and Delay Times in Cyclic Growth

Cyclic bubble growth delay times (after the first nucleation delay) are not found significantly different: 5 ± 3 s for fat versus 7 ± 5 s for muscle. Detachment sizes (last size before bubbles float) are significantly different: 439 ± 52 m for fat versus 213 ± 52 m for muscle (Figure 43), possibly due to the differences in geometric and/or wetting properties of the substrates. The observation that detachment size for bubbles on muscle tissue substrate is significantly less than for fat tissue (Figure 43) is also reflected in Figure 42 showing more smaller bubbles compared to bigger ones on muscle.

Figure 43 : Bubble radius at detachment (left, fat n = 8, muscle n = 7) and delay times between bubbles growing from same nucleation site (right, fat n = 7, muscle n = 6 and multiple n = 6), for muscle and fat tissue substrates.

4.3.2.4 Contact Lines and Contact Angles

Image processing results for contact line and contact angles are shown in Figure 44.

Figure 44 : Example bubble tracking result from custom MatLab code

144

In the five cases where the bubble subsequently detaches the contact line shrinks dramatically during that sequence as expected. Excluding these however, for each individual sequence no clear trend is visible in terms of evolution of the contact line and contact angle. Since the rate of bubble growth is slow, this could be due to the time of observation of each sequence combined to the fact that the quantities discussed here are the apparent contact lines and angles (with a higher associated measurement error, possibly due to microstructures or uneven tissue surface). A more representative measurement would therefore be to superimpose all contact angles and contact lines measured.

Mean contact angle (CA) and contact line (CL) distance evolution overtime from calculated bubble inception time (in the same manner as what was presented for Figure 42) were thus plotted and slopes of the evolution fitted linearly extracted (n = 23). The mean contact angle was shown to decrease over the bubble growth time (푦 = −0.014푡 + 35, G = 0.54), whereas the contact line trend was shown positive (푦 = +0.04푡 + 116) but with only G = 0.1 making this a weak trend.

For a gas bubble growing from a substrate’s cavity, it is expected that the contact angles will shrink as the bubble grows outward if it remains spherical which is the case in our experiment. The question remains as to whether the contact line stays anchored at the mouth of the cavity during growth. All but three of our data regarding CL and CA come from the fat tissue substrate (no difference was immediately obvious between the muscle and fat ones, however not enough data is available to conclude) and fat is supposed to be hydrophobic from the literature [214, 309]. It is expected that a bubble growing on a hydrophobic surface will free itself from the cavity’s mouth and thus its contact line would be shown to increase over its growth time [310]. This is still plausible from our results given the positive slope found for the contact line over bubble growth time, however the weak correlation (퐺 = 0.1) together with the fact that the contact angles measured on fat are consistently well below 90° point to our substrate not being particularly hydrophobic in this case.

4.3.2.5 Multiple Bubbles: competition for dissolved gas and coalescence

A competition for dissolved gas between adjacent bubbles is demonstrated in increased delay times in cyclic growths from the same nucleation site once a bubble has detached (Figure 43), as well as slower growth rates for non-isolated bubbles. The growth rate of bubbles growing 145 close to each other departs for the constrained parabolic law fitting, which is evidence for dissolved gas competition amongst multiple bubbles growing in close proximity. For multiple adjacent bubbles, the 퐺 coefficients for muscle and fat were indeed significantly worse than for single bubbles at 0.84 ± 0.20 and 0.92 ± 0.08 respectively. Interestingly, in almost all cases one bubble of the adjacent multiples seemed to dominate in the competition for dissolved gas from the liquid as demonstrated by a 퐺 closer to the single bubbles’ and the rest of the bubbles showed significant departures from the expected mass diffusion growth curves (with 퐺 as low as 0.38).

4.3.4 Discussion

Results from this study indicate that, by effectively decoupling the effects from tissue gas absorbance rate and tissue surface, it is possible to ascertain that tissue surface also plays a role in the difference observed between fat and muscle bubbles and that this is not solely due to fat tissues absorbing more inert gas as is often quoted in historical physiological papers. The key novelty in this work, however, remains the development of an experimental set-up, as well as specific data processing software, for studying decompression induced bubble growth rate and density on ex vivo tissue surfaces that we consider as “dead” (not metabolizing), with real-time optical imaging acquisition and full temperature control, down to μm resolution. This method can be directly extended to the study of different pressure profiles, gas mixtures, tissues or other substrates, temperature regulations and liquid compositions, etc. In addition, in vitro or in vivo experiments in the small decompression chamber could also be considered with the same imaging and data processing capabilities as described presently. The use of a circular Hough transform based algorithm has the clear advantage of complete robustness if the bubble grows slightly out of the focus plane field, as the principal radius component is picked up through an ordered voting process. In addition, lighting the bubbles from the back permits us to focus the camera in the exact centre of the bubble when first observed as they appear black with a fine white inner circle due to the light set-up chosen. Combined to the facts that the optical acquisition is done automatically from the outside of the chamber and the temperature controlled throughout the experiment, this allows for the most precision of the measurements as the set-up is properly sealed throughout the measurements with samples not even touched (and indeed anecdotally on touching the

146 small decompression chamber without even attempting to open it bubbles were seen to immediately start to float). For multiple bubbles growing in close proximity, it was shown that the 퐺 goodness of fit coefficient for parabolic growth is on average less good than for single bubbles. Interestingly the standard deviation is also significantly increased. This is due to the fact that in most cases of multiple bubble growths there is one bubble that dominates (with higher 퐺 almost comparable to single bubbles) and the other growing significantly slower. This competition for dissolved gas is also demonstrated in the significant bigger delay times between successive bubble cycles growing from the same cavity when multiple bubbles are present (Figure 43). If this competition for dissolved gas which has been demonstrated for instance in heat transfer bubble growth in microgravity experiments [119], from the liquid is correct, it is expected that the average 퐺 coefficient for multiple bubbles’ growth fitting will drop as the distance between two adjacent bubbles decreases or as the number of adjacent bubbles increases. In addition, above a certain distance threshold between bubbles it is expected that the deviation from single bubble behaviour will be negligible. As a very preliminary test for this hypothesis and as a means to generate a hypothesis for further validation studies, taking only the cases where there are only two adjacent bubbles and both are in focus, Figure 45 shows the trend for the 퐺 coefficient against the inverse of the distance between bubble centres in μm. However, the scarcity of available data does not permit us to conclude solidly on this yet and more observations are needed. This is difficult in practice as the position of appearance of the bubbles with respect to the focusing and field of view of the camera is luck and in our experiments, despite having hundreds of bubble sequences only four cases with only two bubbles and both in focus are recorded. Nevertheless, from this preliminary analysis shown in Figure 45, it is expected that above around 100 μm distance between the centres the bubbles then the results will be comparable to single growth bubbles where competition for dissolved gas is negligible, which is reasonable from diffusion lengthscale arguments.

147

Figure 45 : Inverse relation between distance (in μm) between two adjacent bubbles and their average G coefficient.

The multiple bubble behaviour also differs between fat and muscle tissue substrates, with bubbles on muscle growing until merging and the dominant bubble then suddenly increasing its radius (such that the total volume of the merged bubble corresponds to the sum of volumes of the previous bubbles), whereas multiple bubbles on fat grow until they touch each other and coalesce without merging. This was observed in all bubbles seen to grow to the point where they touched each other. In both cases a translational motion (as in [118]) of the bubbles toward each other is often observed with the slower growing bubble moving preferably toward the dominant one. Initial nucleation delays for the different tissue substrates can be extracted from the bubble density observations, reflecting the number of nucleation sites, the growth rate, detachment size, as well as delay between bubble cycles information. Looking at the bubbles which are first generation cycles only, the initial distribution of delay times for fat is found to be 180 ± 106 s post decompression start and for muscle 155 ± 59 s post decompression start. This approach, together with the different multiple bubble behaviour from different tissues, could be used to feed initial parameters and strategies for decompression models in the future.

148

4.3.5 Conclusions

In conclusion, we describe an experimental set up developed for recording bubbles growth rates on tissue surfaces from gas saturated solutions during a reduction of ambient pressure (decompression), as well as bubble density per tissue area. Although the experiment reported here treats the case of air saturated distilled water and ex vivo rabbit muscle and fat tissues from 3 to 0 bar decompression at 1 bar/min, this can be extended to other liquid or gas properties and composition, temperatures, decompression profile and nucleation site (tissue surface structure and hydrophobicity).

Muscle and fat tissues from rabbits show significantly different bubble densities with the same decompression profile (3 to 0 bar) but similar growth rates once the bubble has reached a critical size. Heterogeneous nucleation is observed from preferential sites on the tissue substrate, where the bubbles grow, detach and a new bubble forms from the same site in turn. Bubble density, the number of bubbles per unit area (measured at 30 min post decompression start), is found significantly different between the two tissue types with more sites on fat observed. Cyclic bubble growth delay times (after the first nucleation delay) are not found significantly different: 5 ± 3 s for fat versus 7 ± 5 s for fat, nor are the initial delay distributions. Detachment sizes (last size before bubbles float) are significantly different: 439 ± 52 μm for fat versus 213 ± 52 μm for muscle. Finally, a competition for dissolved gas between adjacent bubbles is demonstrated in increased subsequent delay times as well as slower growth rates for non-isolated bubbles.

The role of the tissue substrate, decoupled from the absorption of gas with this new experimental set-up, is therefore demonstrated to also play a role in bubble growth, merging/non-merging behaviour, as well as detachment from decompression.

149

150

Chapter 5 – Development of new endpoint evaluation using ultrasound imaging

5.1 New bubble counting methodology statistical comparison11

This section deals with the development of a new methodology for evaluating the bubbles observed with echocardiography post dive, which resulted in a publication in collaboration with our clinical collaborators11. My direct contribution within this work was to develop and then implement the statistical methods to evaluate the reliability (repeatability for a same person, and agreement between subgroups of raters) in order to evaluate this method (sections 5.1.2.1 Statistical Methods and 5.1.3 Results). The method assessed was devised by Dr Peter Germonpré, MD, of Centre for Hyperbaric Oxygen Therapy, Military Hospital, Brussels, Belgium, and the full paper is reproduced here with permission.

5.1.1 Introduction

Underwater diving on compressed air or other breathing gases exposes the diver to so-called ‘decompression stress’, caused by the release of nitrogen and/or other inert gases from the body tissues during and after ascent from depth, resulting in bubbles forming in tissues and (more commonly observable) in blood. In order to minimise this stress and decrease the risk of decompression sickness (DCS), decompression algorithms, summarised in dive tables or incorporated into dive computers, have been developed. These algorithms are not completely successful in the avoidance of every instance of DCS and, to this day, a major research effort is directed to identifying factors and interventions (pre dive, during the dive and post dive) that could make decompression safer [29].

11 Adapted from P. Germonpré, V. Papadopoulou et al. Diving and Hyperbaric Medicine; 2014 Mar;44(1):5-13. 151

Evaluation of these algorithms and of the efficacy or inefficacy of other preventive measures has been done primarily on the basis of the presence or absence of clinical symptoms of DCS, as well as on the detection of bubbles in the vascular system using Doppler ultrasonic bubble detectors. Doppler bubble ‘grades’ were first defined by Spencer et al. in 1974, and classified into 5 grades (0 to 4), depending on the number of acoustic bubble signals audible in the precordial region [311]:

Grade 0 – Complete lack of bubbles Grade 1 – Occasional bubble signal, vast majority of cardiac cycles bubble-free Grade 2 – Many, but less than half, of cardiac cycles contain bubbles, singly or in groups Grade 3 – All cardiac cycles contain bubbles in showers, but not overriding heart signals Grade 4 – Bubbles sounding continuously during systole and diastole, overriding amplitude of normal heart signals

Table 5: Spencer grading system for VGE on echocardiograms post dive

In 1976, Kisman and Masurel defined a scale using three parameters (frequency, amplitude and duration) allowing for more precise classification but rendering acquisition and evaluation more complicated [11, 312]. Both these scales require a skilled, experienced Doppler technician in order to be reproducible [313, 314]. In 2004, Divers Alert Network (DAN) Europe Research proposed a simplified ‘bubble score’, distinguishing only low, medium, high and very high bubble grades based on precordial Doppler, but this scale has not been widely adopted by others [315, 316]. Modifications of the original Spencer scale have likewise been proposed, resulting in the ‘Expanded Spencer Scale’, with a larger number of categories and thus a more incremental grading [315, 316]. Whilst the original Spencer scale has been by far the most frequently used in diving research, the Kisman-Masurel scale has been preferred for large, well-controlled, laboratory decompression research studies, and an association between bubble grade and risk for decompression sickness has been developed that can equally be used for the Spencer scale [311]. Generally, it is accepted that the higher the number of bubbles detected precordially, the higher the statistical risk for DCS after a dive [11, 314, 317].

152

Using echocardiography, Eftedal and Brubakk in 1997 proposed a bubble score of six grades based on visual analysis of 2D precordial echo images [318]:

Grade 0 – No observable bubbles Grade 1 – Occasional bubbles Grade 2 – At least one bubble every four cardiac cycles Grade 3 – At least one bubble every cardiac cycle Grade 4 – At least one bubble per cm² in every image Grade 5 – ‘White-out’, single bubbles cannot be discriminated

Table 6: Eftedal and Brubakk system for VGE grading post dive on echocardiograms

This bubble score allows a semi-quantitative evaluation in a reproducible manner, with minimal intra- and inter-observer variability. However, the scoring system as proposed does not discriminate well in the medium range of bubble scoring, with a large jump between grade 3 and grade 4, making this score less adapted for the evaluation of low to medium levels of decompression stress (classifying into either ‘low’ or ‘severe’). Also, the use of echocardiography made this method less practical for deployment in real-life diving situations (e.g., on a dive boat with a humid, sometimes cold environment and possible lack of AC power). Only recently have good-quality, portable echocardiographs become available, that make on-site evaluation (at the waterfront) possible, by visualising decompression VGE. The use of ‘harmonic imaging’ (HI) decreases noise in the cardiac cavities, and Color Map application (‘gold’ setting instead of standard ‘grey’) provides better image contrast [319, 320]. Thus, the detection of Venous Gas Emboli (VGE) in divers’ heart cavities and large veins is easier and visualisation of smaller VGE than were detectable by older echography machines is possible [321]. Of note, this use of HI improves the signal-to-noise ratio and increases contrast, but does not aim to make VGE oscillate to emit their own harmonic frequencies, as much lower scanning frequencies would be needed for this to happen [171, 322-324]. For a useful review of HI the reader is referred to references [319] and [320].

153

Here, we describe a newly developed method of evaluation of decompression-induced VGE, using transthoracic 2D echocardiography, which may offer significant advantages compared to current methods.

5.1.2 Methods

A standardised technique for evaluation of decompression stress by means of counting the number of VGE in the right heart chambers (venous blood) is described, using a portable echocardiography device, with hard-disk recording and a posteriori (off-line) evaluation of cardiac images. The technique was developed using a Vivid-I portable echograph (GE Healthcare, UK) and subsequently applied successfully using a Vivid 7 echograph (GE Healthcare, UK), both in a controlled environment (beside a swimming pool) and in the field (dressing room of a Belgian quarry dive site).

A GE 3S-RS sector array ultrasound probe (GE Healthcare, UK) is used; the machine is used in harmonic imaging mode (2.0/4.0 MHz). A four-chamber view is obtained by placing the probe at the level of the left fifth intercostal space. It is necessary to modify the standard four- chamber view by rotating the probe slightly ventrally (in the direction of the xyphoid process) so the right atrium and ventricle can be fully visualised. Three ‘landmark points’ are identified to aid proper positioning of the ultrasound probe: both transsections of the tricuspid ring and the top of the right ventricle should be visible in the image (Figure 46). A series of at least 15 cardiac cycles are recorded onto the internal hard disk of the echograph while keeping the probe immobile. With practice, each recording can be done in less than 3 minutes (positioning of the diver, attachment of three ECG electrodes, obtaining a good view, recording, detaching the electrodes), allowing for serial measurements on up to 10 divers within a 30-minute interval between measurements of the same diver. At the completion of the measuring period, all videos are saved onto external hard disk or USB thumb drive in the ‘wmv’ format (Windows Media Video, at 30 frames per second), for which GE Healthcare provided a proprietary video player (MPEGVue Player).

At a later stage, the recordings stored on the portable hard disk are reviewed using the MPEGVue software (GE Healthcare, UK), which allows for easy patient and examination selection, frame-by-frame advancing of the video frames using the keyboard arrow keys and freezing of the video frames while maintaining good still-image quality. First, the pre-dive 154 echography loops are reviewed in order to identify intra-cardiac structures that may mimic VGE (e.g., papillary muscles, valve leaflets, Chiari network, Valsalva sinus). Then, the post- dive echography is reviewed and played in a loop at real-time speed in order to rapidly assess the presence or not of circulating bubbles. In cases where bubbles are seen, a formal bubble counting procedure is performed. Using the pause button, the loop is frozen at the start, and then with the forwards and backwards buttons, an image frame is selected in end- diastolic/proto-systolic position (where the tricuspid valve leaflets are fully opened and almost invisible) (Figure 47) and bubbles are counted in both the right atrium and ventricle (Figure 48). In case the chosen view does not contain any bubbles, but bubbles are clearly present in the heart cycle, the forwards and backwards buttons are used to select another frame, within two to three frames of the frame originally chosen. Ten consecutive frames are analysed and the bubble count is averaged over these 10 frames.

Figure 46: Landmark structures in the right heart echography image: the upper circle identifies the ‘top’ of the right ventricle (RV) while the lower two circles identify the section through the tricuspid annulus on either side of the right atrium and constitute the ‘upper’ border of the RA. (N.B., echocardiograph images are inverted).

155

Figure 47: Choice of frame to analyse: the three landmark circles are drawn as in Figure 46. The frame chosen for analysis is indicated by the red marker on the electrocardiography trace (marked by the small green circle, bottom right). Both leaflets of the tricuspid valve are fully open and visible against the ventricular wall (points of green arrows); the right atrium and ventricle form a single cavity.

Figure 48: Bubble counting: bubble signals are identified as bright spots and counted individually; tricuspid valve leaflets and other fixed structures (e.g., papillary muscles in the top of the right ventricle) are not counted.

156

The technique was developed for use during a series of standardised test dives organised by DAN Europe Research (Roseto, Italy and Brussels, Belgium), in an indoor swimming pool of 34 metres’ fresh water (mfw) depth (Nemo33, Brussels, Belgium). The dives were designed to evaluate the effect of several pre-dive interventions on the number of VGE post dive. For this purpose, each diver performed one (identical) dive per week, to 33 mfw for 20 minutes. This ‘standard’ dive was performed at least three times under ‘normal’ conditions, and several times under ‘experimental’ conditions, when the effects of several methods of preconditioning were measured. The order of the experimental dives was randomised. Each diver was evaluated with, among other tests, precordial echocardiography at three time points: before the dive, at 30 minutes and at 90 minutes after surfacing. The study was approved by the Academic Bioethical Committee of the Free University of Brussels (CE/2008/66); all divers were unpaid volunteers who provided written informed consent.

In order to verify the internal (intra-rater) and external (inter-rater) consistency of this frame- based counting method, nine observers were asked to perform analysis of the same set of images. Three were trained cardiologists, at various times involved in diving research performed by DAN Europe. All had performed one or more image acquisition sessions during the experimental pool dives. Three were medical doctors from the Centre of Hyperbaric Oxygen Therapy of the Military Hospital Brussels, who had no formal cardiology training but were present during some or all of the diving experiments, and had some experience in viewing echocardiographic images. The third group consisted of DAN Europe researchers or certified hyperbaric technicians (CHT) from the Centre of Hyperbaric Oxygen Therapy, who had various degrees of paramedical training, allowing them to identify the major intra-cardiac structures after some instruction. All received written instructions detailing the evaluation procedure (and containing the same pictures as in this report) and a short period of hands-on training in the use of the MPEGVue software, which is simple and intuitive to use.

First, a test was administered to verify the reliability and repeatability of the VGE counting by itself. A set of 50 still-frame images was presented for static bubble counting. These images were extracted by the authors from the available video loops, and chosen so as to represent a mix of better- and worse-quality images containing between 0 and 40 VGE signals. Images were presented as a Microsoft PowerPoint presentation. No identifying elements (such as name, birthdate, acquisition date) were displayed on the images, only the

157 slide number. No time limit was given for viewing the slides. Unknown to the test persons, several of the slides were in fact identical but spread out randomly over the presentation. Then, a selection of 20 post-dive video sequences were presented, together with their baseline predive echocardiographic loop (no bubbles present) and the observers were asked to evaluate these video loops, using first the Eftedal and Brubakk score, then using frame-based counting as described above.

As there is no way to determine the exact number of VGE in the images, obviously a true ‘gold standard’ cannot be determined. The need to set a standard by which to compare the data from this study prompted us to define a ‘reference score’ as the number of visible bubbles in each image and video loop, agreed on by a priori consensus by the main authors of the study.

5.1.2.1 Statistical methods

Internal consistency was verified on the static images; external consistency was verified on the static and video images with both scoring systems, using the following statistical methods.

Eftedal and Brubakk score

The weighted kappa statistic was chosen to evaluate the inter-rater agreement, in accordance with the discussion on the appropriateness of statistical methods to this effect by Sawatzky [313]. Cohen’s kappa (κ) statistic is used to calculate the coefficient of agreement between raters for nominal grades where the outcome of agreement is binary: either agreement or disagreement [325-327]. For ordinal scales, the degree of agreement should be taken into account and this is done using the weighted kappa statistic instead. Both the kappa and weighted kappa are completely corrected for chance agreement [325]. The weights chosen to weight disagreements were defined in the same manner as the original Eftedal and Brubakk method to allow direct comparison. Since the data are ordinal (but not continuous) for the Brubakk and Eftedal method, a disagreement is ‘stronger’ if one rater assigns a score of 4 and another a score of 1, compared to 1 and 2 respectively. This is taken into account by using

158 weights for characterising the degree of disagreement. In the usual contingency tables for two raters, the weights were specified as:

|푖 − 푗| 휔 = 푖푗 푘 − 1 (37) where i and j index the rows and columns and k is the maximum number of possible ratings. The weighted kappa is then calculated from the proportional observed and expected agreements [325, 328]:

푘 푘 1 푃표(휔) = ∑ ∑ 휔 푓 푘 푖푗 푖푗 푖=1 푗=1 (38) and

푘 푘 1 푃푒(휔) = ∑ ∑ 휔 푟 푐 푘2 푖푗 푖푗 푗 푖=1 푗=1 (39)

where 푓푖푗 is the number of recordings graded i by one rater and j by the other, 푟푖 is the row total for grade i and 푐푗 is the column total for grade j, such that:

푃표(휔) − 푃푒(휔) 푤푒푖푔ℎ푡푒푑 푘푎푝푝푎 = 1 − 푃푒(휔) (40)

The kappa-statistic measure is a value between -1 and 1, with 0 corresponding to the value expected by chance and 1 perfect agreement. The interpretation of the values as suggested by Landis and Koch are given as [329, 330]:

159

below 0.00 – Poor 0.00–0.20 – Slight 0.21–0.40 – Fair 0.41–0.60 – Moderate 0.61–0.80 – Substantial 0.81–1.00 – Almost perfect

Table 7: Proposed interpretation for Kappa Statistic values

Frame-based counting method

For the frame-based counting method, both on still images and on the average over 10 video frames, the data are also ordinal but this time continuous (video) or discrete (units of bubbles). The weighted kappa statistic cannot be used for continuous variables [331]. Therefore, another statistical test has to be chosen. For continuous data the intraclass correlation coefficient should be used as a measure of reliability, or Bland-Altman plots for limits of agreement and bias [331, 332]. The intra-class correlation coefficient or ICC gives a measure of the proportion of total variance due to the difference between raters by penalising systematic error. For ordinal data, the intra-class correlation coefficient is comparable to the weighted kappa statistic if quadratic weights are used, which is why both weighted kappas (linear as in Sawatzky, and quadratic for comparing with the ICC) are quoted here [313, 333]. Note that it is exactly equivalent only for uniform marginal distributions [332, 334]. The ICC scale goes from 0 to 1, with 1 representing perfect agreement and 0 no agreement. The Bland- Altman plot displays for two assessors (or groups of assessors) the difference for each assessment against the mean of each assessment [328, 335]. The confidence interval is also displayed, calculated as the 95% percentiles such that the upper and lower bounds are given by:

Means of differences ± 1.96 (std of differences) ( 41 ) As such, the Bland-Altman plot shows any bias and the limits of agreement between two raters.

160

Intra-rater reliability (internal consistency)

The intra-rater reliability was assessed on the still-images test for the repeated images by the Wilcoxon signed-rank test, calculating the Spearman rho (rank correlation coefficient ρ) for every rater on the repeated images counts (taking the maximum discrepancy for the one image repeated three times). The value of ρ lies between -1 and 1, a higher number indicating a better reliability. The calculation of the weighted kappa statistic and ICC was performed offline using the standard statistical package Stata (StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX: StataCorp LP). All other data processing and plotting was done by calculating the appropriate values offline as defined above directly in the commercial software package MatLab (MATLAB 6.1, The MathWorks Inc., Natick, MA, 2000).

Rater Category Spearman rho 1 C 0.9733 2 C 0.9487 3 C 0.2052 4 MD 0.9211 5 MD 0.7632 6 MD 0.9211 7 O 0.7632 8 O 0.9747 9 O 0.8922

Table 8: Static images bubble counting – identical image pairs scores; Spearman ρ between raters and a reference score (see text); all comparisons non-significant (Wilcoxon test-retest p>0.05); C – cardiologist, MD – physician, O – other (paramedic or hyperbaric chamber attendant)

161

5.1.3 Results

After some practice runs with the frame-based method, all observers reported bubble counting to be relatively easy and rapid, although the process of scrolling through video files was found to be somewhat tedious and slow (approximately 5 minutes for a video file evaluation). The static images were less confidently scored because, as the raters reported, no video images were available to help discriminate between intracardiac structures and VGE. However, the number of bubbles counted was not significantly different between observers (absolute number of bubbles 0 to 40 bubbles). As expected, a larger standard deviation was observed for larger bubble numbers. The ICC between the reference score and all raters was 0.96 (95% confidence interval (CI) from 0.92 to 0.99).

Calculated differences in scoring for identical image pairs (intra-rater or internal consistency) were non-significant (Wilcoxon test-retest, P > 0.05) with excellent Spearman ρ (0.76 to 0.97) except for one cardiologist, rater C3 (ρ = 0.21, Table 8). Further analysis showed that this observer consistently scored approximately 5 bubbles higher than the average, suggesting that a systematic error was present (see Bland-Altman plot, Figure 49). However, even in the case of this cardiologist with lower Spearman ρ, the Wilcoxon test retest P-value showed that the differences in the test-retest counts were non-significant.

For the video sequences, the Eftedal and Brubakk scoring gave a weighted kappa of κ = 0.5815 with linear weights and κ = 0.7634 with quadratic weights, which shows a moderately good external consistency. It was found to be slightly lower than reported in the original publication (κ = 0.6796 using linear weights) [318]; this may be a reflection of our study design testing and how easy the grading methods are to learn (use of non-expert raters with only written instructions). As indicated in the methods section, all raters received only minimal instructions in the various methods: a three-page document and a short hands-on training session on the use of the video player software. Therefore, the lower external consistency may well reflect the lesser experience in grading according to this score, as none of the nine raters had ever performed an Eftedal and Brubakk scoring before. The ICC for the Eftedal and Brubakk scoring gives 0.79 (95% CI 0.54 to 1.05); as this method is similar to the weighted κ with quadratic weights, it shows a very good inter-rater agreement.

Frame-based counting gave a higher external consistency, with an ICC of 0.84 (95% CI 0.77 to 0.92). There was no significant difference between all observers and the reference score

162

(see Bland-Altman plot, Figure 50); however, here again, the same cardiologist scored consistently approximately 5 bubbles higher on every occasion.

Figure 49: Bland-Altman plot showing systematic over-estimating by Cardiologist 3 as compared to the mean number of VGE counted by all others; X-axis: number of VGE in the image; Y-axis: difference of count vs. mean; horizontal lines – 95% CI as 1.96 std differences; LoA – limits of agreement

Figure 50: Bland-Altman plot showing the good consistency between reference score and all observers for frame-based counting in the video sequences; X-axis: number of VGE in the video sequences (average of 10 frames); Y-axis: difference of count vs. mean; horizontal lines – 95% CI as 1.96 std differences; LoA – limits of agreement

163

5.1.4 Discussion

(Semi-)quantitative determination of VGE is an important, if not still the only tool available for evaluation of diving decompression stress. Currently used methods suffer from either the necessity of highly skilled observers, a complicated evaluation method (Spencer and Kisman- Masurel scales) or a semi-quantitative visual evaluation that fails to discriminate well in the mid-range of VGE (Eftedal and Brubakk score), exactly the range that most interventions to improve decompression safety for recreational divers would act upon. Also, bubble counting takes place only at certain points in time after the dive, and the accuracy of estimating the total bubble load is dependent on the number of measurements and their timing. One method of estimating the bubble load out of a number of discrete bubble evaluations is the Kisman integrated severity score (KISS), which integrates bubble grades from a number of observations over a given time period into a single value; it can be considered an estimate of the ‘area under the bubble grade curve’, and is a relative value that can be used for comparative purposes [336-338].

Using frame-based counting, a continuous-scale (more quantitative) evaluation of VGE presence can be done in a relatively quick, easy way, with good reproducibility. Using the bubble counts for 10 consecutive frames allows for small beat-to-beat variations in bubble numbers to be averaged out. A current drawback is that bubble counting must be done manually at a later stage, which requires additional steps (exporting the video loops in MPEGVue format) and takes some time for counting. Thus, it is not real-time analysis. However, taking into account the echogenicity of the different surrounding structures and using intelligent learning algorithms, computerised automatic counting may become possible. This would permit real-time and continuous counting of VGE, and thus make VGE evaluation independent of the timing of observations after the dive. These algorithms are currently under development [136, 138, 339].

As 2D echocardiography permits viewing the cardiac cavities in a single plane only, the choice of plane may be of some importance. The standard four-chamber view, as used in echocardiography, shows only the basal part of the right ventricle, with the top of the right ventricular cavity out of view. This is not a problem in cardiac evaluation, as most emphasis lies on the morphology and function of the left atrium and ventricle, but may obscure significant parts of the right heart cavities, where VGE are primarily visible after the dive. To overcome this, the method described requires slight tilting of the echo probe to point more in 164 the direction of the xyphoid region, permitting identification of the three landmarks: the top of the right ventricle, the tricuspid ring and the left and right tricuspid valve leaflet bases, in order to maximally expose the right heart cavities (Figure 46).

The selection of the freeze frame where counting will be done is somewhat arbitrary, but based on the following considerations:

 The end-diastolic/proto-systolic time point is when atrial contraction has finished and ventricular contraction has yet to begin. This is the moment in the cardiac cycle when there is the least flow of blood. Although small areas of turbulence cannot be ruled out, there is at least no rapid movement driven by cardiac contraction.  It is also the moment when the tricuspid valve leaflets are fully open and almost invisible, making the right atrium and ventricle into a single blood-filled cavity; this decreases the chance of erroneously interpreting valve leaflets as bubble signals.  This moment is identified easily using the electrocardiographic trace, when recorded with the images.

Although it may be possible theoretically to analyse other frames in the cardiac cycle, these considerations make it unlikely that a better estimation of the number of bubbles might be obtained. In any case, it is important to count the same frame consistently.

Dynamic evaluation such as the Eftedal and Brubakk method seems to slightly over-estimate VGE numbers as compared to actual counting on freeze frames. This can be explained by the fact that vortices of blood exist both in the atrium and ventricle, by which VGE may be swept several times through the plane of vision [340, 341]. These blood-flow patterns account for the fact that in some instances, the ‘correct’ freeze frame chosen for frame-based counting does not show any VGE at all, whereas the previous or next frames do show a significant number (up to 9 or 10) VGE. The procedure therefore allows choosing a frame slightly ‘off’ if there are obviously VGE in the heart cycle but none can be seen in the initially chosen frame. With automated computerised counting, it will be possible, using three to five frames around the optimal frame, to eventually average out these turbulence effects. Currently, the manual method is too slow to reasonably permit counting of more than 10 to 20 frames in a video loop, as a certain degree of ‘observer fatigue’ eventually sets in.

The counting method described here makes use of a proprietary video file player on the PC (MPEGVue) which is offered as a package by the echograph’s manufacturer (GE). This 165 offers the possibility of viewing echocardiography video files off-line on any Windows PC while offering an easy patient selection menu and the possibility to smoothly step forwards and backwards through the video file, making frame-accurate selection of the images possible. Although a large range of video-playing software that can play back ‘wmv’ video files on a PC is available, none of them offer this frame-accurate playback. The major drawback here is that the MPEGVue videoplayer can only play back files if the file structure is organised in a certain way – in practical terms, it limits the application to using GE echographs for acquisition and storage of the videos. All of those echographs offer MPEGVue export of the digital (DICOM) files, and once in the MPEGVue format, video files can be shared using either USB disk or sent by e-mail, with the player installation files added to the export package. Automated software will not suffer from this limitation, as it will be able to digest the individual frames of a video stream or file using proprietary software, e.g., MatLab software (MathWorks, Natick, MA, USA).

The inter-rater agreement for frame-based counting is high (ICC of 0.84), indicating there is no major difference between the individual observers and the reference score. This would permit pooling of data from different observers within the same experimental data set. As the VGE counts are an ordinal and continuous variable, mean and average VGE numbers can be calculated, which represents an obvious advantage over the use of discrete variables such as bubble grade scores for evaluating decompression stress. However, the almost perfect (ICC 0.96) intra-rater consistency for this method means that having the same assessor count VGEs for a set of experimental data would give extremely reliable results with regards to the evolution of VGE numbers post dive. Of course, it will be necessary to verify the (intra-rater) consistency of the computer automated counting software which is being developed. If confirmed, this software could be used either for off-line analysis of large numbers of files or, perhaps, directly on an ultrasound scanner (real-time evaluation). At present, the time- consuming process of counting individual bubbles and moving back and forth between frames to discriminate bubbles from their paths and movement prohibits large-scale use of the method.

It has been correctly pointed out that newer echocardiographic techniques are able to detect much smaller bubbles and that, as a result, it is impossible to compare published research using counted bubbles on echography unless exactly the same settings are used. Specifically, Eftedal and Brubakk scores will be impossible to compare among different studies, and it will

166 be impossible to compare the effect of (pre-) diving interventions on VGE production with previous data from similar dives because of this. Recent case reports have indeed described divers with Eftedal and Brubakk grade 5 cardiac echograms (initially thought to be almost impossible without resulting severe DCS), without any symptoms of DCS [342]. This is undoubtedly a result of the better spatial resolution of modern echography, and the use of second harmonics imaging [321].

The same applies for frame-based bubble counting; it is important to obtain baseline, control dive and postintervention images on the same group of divers. However, the continuous-scale nature of this method will permit a quantitative evaluation of the effect of the intervention on VGE production. This way, even if the echographic method per se changes and becomes more sensitive, the relative effect observed in different studies may be compared.

Finally, using echocardiography, it may also be possible to evaluate (de)hydration state (by the degree of respiratory collapse of the inferior vena cava, IVC) and, in some subjects, decompression bubbles may even be detected in the IVC and the portal veins [343-345]. Incorporation of this information may provide additional insights into the influence of factors unrelated to the dive profile itself on the production of VGE after the dive. Using solely the degree of VGE after a dive as a measure of dive profile safety without at least trying to standardize these individual (diver-related) factors that may make a diver, either constitutionally or temporarily, less or more prone to the production and liberation of VGE after a dive, disregards a mass of scientific information already available on this subject [43, 70, 71, 346, 347]. The presence of VGE in the left cardiac cavities after a dive, be it by passage through a patent foramen ovale or through pulmonary arteriovenous shunts, may indicate a higher risk for cerebral or high-spinal DCS in the individual diver [185, 348]. This may guide a decision as to whether a particular diver should be excluded from further participation in diving studies, especially if high risk.

5.1.5 Conclusions

As opposed to existing methods of evaluation, a frame-based counting method permits the investigator to define bubbles as a continuous variable, allowing more flexible and powerful statistical evaluation of the presence of VGE as an indicator of decompression stress. The method presented here shows excellent inter- and intra-rater consistencies, which can be 167 achieved with minimal training by non-experts. Because of the linear, continuous-scale nature of the evaluation, a better discrimination of VGE levels can be achieved in the important intermediate range of bubble load. Therefore, the method seems well suited for use in interventional human diving experiments, where it is ethically impossible to subject volunteer divers to dive profiles generating extreme bubble grades. Moreover, the method is suitable for the development of automated counting software.

168

5.2 Use on real dive data from field experiments

The bubble counting method developed in Section 5.1 was used to analyse echocardiographs in two sets of field experiments. The first was in the setting of collaborative experiments on scuba divers led by Dr Miroslav Rozloznik and taking place in the Austrian lake Attersee. The second implementation was during collaborative experiments led by myself, with the help of Dr Sigrid Theunissen, in the NEMO33 deep diving pool in Brussels. The aims of both field experiments go beyond what is presented in the thesis. The scope of this section is therefore limited to the echocardiography findings with respect to the new endpoint evaluation method introduced in Section 5.1.

5.2.1 Attersee experiments

5.2.1.1 Background

The Attersee lake in Austria has a year round water temperature of about 4°C. The main aim of the experiment led by Dr Miroslav Rozloznik was to investigate the cognitive function of divers for cold water between Air and Trimix dives with matched calculated narcotic profile, as assessed by a battery of psychometric tests implemented on a Mares dive computer (collaboration between Prof. Balestra and the company Mares in the PHYPODE project) and another cognitive arousal test, the CFFF (Critical Flicker Fusion Frequency) device. The results could then be compared to previous data in warm water. Due to the above aim, as well as extensive technical diving experience of the volunteer divers, the profiles for the two dives were as follows:

 Air dive (79% nitrogen, 21% oxygen) to 30 meters depth for 20min, with a total dive time of 35 minutes, with no decompression stops.  Trimix (10%O2, 55%He and 35%N2) dive to 70 meters depth for 20min, with for a total dive time of 100 minutes, including gas switching and decompression stops as follows: Nitrox 50 from 21m to 6m depth (decompression stops of 9min at 21m 6min at 18m, 2min at 15m, 2min at 12m and 4min at 9m), then pure oxygen with air breaks from 6m to the surface (decompression stop of 23min at 6m).

169

Six divers, aged 42±7 years old, BMI 26.6±1.9, were measured pre and post-dive for several physiological parameters, including echocardiography. In all divers, specific venous gas emboli (VGE) videos were acquired following the protocol developed in Section 5.1, at the 30min post dive point. For two divers, it was possible to measure divers at 5 different time points post dive and look for evolution in the number of bubbles observed (not possible for all divers due to experimental constraints).

Indeed, we hypothesize [349] that the evolution of VGE in time post-dive might be related to the evolution of other physiological parameters related to hydration, exercise or inflammation-induced vascular wall dysfunction. This experiment served as a test for how to implement in the field, in conjunction with numerous other parameters monitored, multiple- timepoints VGE evaluation, in preparation for a future experiment investigating this specifically.

5.2.1.2 Results and Discussion

Trimix dives

No bubbles were observed for 4/6 subjects, evaluated at 30min post dive and 90min post dive. In one subject 0.13 bubbles were observed at 35min, then 0.11 at 80min post dive. In the last subject, 23.25 bubbles were observed 35min post dive (no later measurement).

The lack of bubbles in most subjects is not surprising giving the extensive pure oxygen last decompression stop, which would likely dissolve any bubbles forming. Nevertheless, it is interesting to see that in one subject the number of bubbles observed was significantly higher than the rest of the group for the same dive profile. This illustrates how different people will yield different results even for the same exposure.

170

Air dives

The bubbles observed on echocardiograms in the six divers are shown in Figure 51, where multiple videos for each timepoint were averaged to give the final results (66 videos were analysed in total). From the evolution of these in time, it is clear that not all divers follow the same trend and the bubble peak can be reached at different times post dive for different people. It is then clearly interesting to develop a regular post dive monitoring protocol so as to best investigate bubble links to other physiological parameters [134].

Figure 51: Bubble counting score for all 6 subjects after the Attersee air dives using the new counting methodology (as described in Section 5.1)

5.2.2 NEMO33 experiments

5.2.2.1 Aims and constraints

The main aim of this experiment was to assist Prof. Karapantsios’ group to assess a new technique for vascular emboli detection based on electrical resistance measurements developed with the European Space Agency [350]. These were coordinated by myself with the help of Dr Sigrid Theunissen and under the supervision of Prof. Balestra. Since these require a lot of organization, additional teams and aims were incorporated to maximize

171 output. These are beyond the scope of the work presented in this thesis, where only the evolution of VGE post-dive will be discussed. The standardized NEMO dive, 33meters for 20min on air, was chosen for these experiments so that results could be taken into the context of years of previous studies from Prof. Balestra’s group. In order to take advantage of these experiments for using the new scoring system developed in Section 5.1, very frequent timepoints of measurements post-dive were favoured. The different physiological parameters measured post-dive are usually done only twice post dive at an hour interval for practical reasons. Instead, we chose to measure much more frequently with the overall goal to look for correlations in the evolution of all these parameters together. This was also key to assessing the new technique developed by Prof. Karapantsios’ group, as their measurements could likely depend on vascular volume (hydration), bubbles, Flow Mediated Dilation (FMD), etc.

Very briefly the protocol was therefore based on all the following:

 Collaboration with Prof. Karapantsios: new bubble measurement technique to be correlated with standardized VGE scoring on echocardiograms, as well as other physiological parameters potentially useful for interpretation (FMD, hydration, etc)  My additional aim, devised with help from Dr Tang, Dr Eckersley and Dr David Cosgrove: try for the first time a PI mode on echocardiography and see whether additional information about bubble size could be evaluated qualitatively  Dr Antoine Boutros, MD: air consumption during the dive correlations to BMI, VO2max and CFFF  Student under my supervision, Nicolas Renne: FMD evolution post dive and grip test (simple force measure) correlations to other parameters

5.2.2.2 Experimental plan

The plan was formulated so that we could measure 10 divers (5 pairs so that divers dove in twos for security, Table 9) on each day over a weekend so that the team from Greece could join. Due to the pool availability (late evenings only on weekend) and measurement stations to correlate, we could have 5 timepoints in total: one pre-dive and 4 post-dive starting at 14min after the dive to leave time for the divers to change, then every 35min thereafter, as per Table 10 and Table 11.

172

Group 1 Diver 1 + Diver 2 Group 2 Diver 3 + Diver 4 Group 3 Diver 5 + Diver 6 Group 4 Diver 7 + Diver 8 Group 5 Diver 9 + Diver 10

Table 9: Planned organisation of subject groups for each day, NEMO experiments

Urine surface tension (only for postdive1); Station A 7min impedancemetry; CFFF; grip test Station B Echography 7min Station C Greek team 7min Station D FMD 7min

Table 10: Rotation of measurement stations for every timepoint, NEMO experiments

Pre-dive Before dive Postdive1 Post-dive (14min post water exit) Postdive2 35min later than Postdive1, 49min post water exit Postdive3 70min later than Postdive1, 84min post water exit Postdive4 105min later than Postdive1, 119min post water exit

Table 11: Timepoints of measurement pre and post dive for the NEMO experiments; the times stated correspond to the start of Station A, then +7min for following stations

173

Figure 52: Example schedule of pre and post dive measurements for one group during the NEMO experiments

Additional considerations

A support crew for diving safety was also present, with 3 divers in the water at set depths at all times during the diving portion of the experiment, 2 other divers outside the water monitoring volunteers and schedules, and a hyperbaric qualified medical doctor present each day with open communication line to the hyperbaric chamber. Taking all the above into consideration, the experimenters were recording back to back measurements for 4hours on each station, totalling 39 people to coordinate per day.

174

5.2.2.3 Bubble counting results

Over the two days of experiments, a total of 17 dives were performed: 10 divers on the first day and 7 divers on the second day (due to last minute cancellations). All measurements for all measurement time points and all measurement stations were acquired with no missing values.

Regarding the echocardiography measurements, a total of 136 videos were acquired for bubble counting, as per:

17 푑푖푣푒푟푠 × 4 푡푖푚푒푝표푖푛푡푠 × 2 퐵푚표푑푒 푟푒푐표푟푑푖푛푔푠 (푟푒푠푡 푎푛푑 푓푙푒푥). ( 42 )

Figure 53 shows the results for all 10 divers of the first day from the “rest” videos (due to time constraints the 7 dives of the second day are not yet analyzed). The scoring was done as per the method presented in Section 5.1.

175

: Evolution of bubble score post dive, NEMO experiments (10 experiments post NEMO dive, subjects) : Evolution of bubble score 53

176 Figure

5.2.2.3 Discussion

The evolution of VGE post dive clearly differs between individuals even when they have undergone the same dive profile. In particular, in divers for whom significant circulating bubbles are visible, the peak score varies and is observed at earlier or later time post dive (Figure 53). For better illustration, Figure 54 presents a frame of the videos analysed for two different subject, showing how subject a peaks around Postdive2 whereas subject b peaks around Postdive1.

Figure 54: Example of echocardiography frame for two subjects, a) and b), at all four post dive measurement time points; subjects a and b correspond respectively to subjects 4 and 8 in Figure 53.

Future work on this data will look at correlations between the bubble score and other physiological parameters, including hydration, for potential explanations regarding off- gassing dynamics. Since the evolution post dive follows different trends for different people, it is clearly advantageous to have a scoring system that allows for more “continuous” evaluation post dive. In this regards, taking multiple echocardiograms and using the scoring developed in Section 5.1 works well, but the manual counting is very time-consuming so semi-automatizing this for experimenter ease and speed is needed.

To give an idea of the time taken to analyse each video, once all frames from each video were extracted to a folder to allow scrolling between them, and once excel templates for inputting all values were created, it took on average 8.1 ± 3.8 minutes to implement the scoring (average calculated from the 40 NEMO videos analysed). As expected videos with no or

177 extremely few visible VGE were faster to score (less than 4 min) and the videos with the most bubbles took the longest (above 12 min).

5.2.3 Conclusion

The method developed in Section 5.1 for assessing VGE post-dive using echocardiography was applied to data from new field experiments, analysing over 100 videos. Once each frame of each video was extracted to an individual image to allow scrolling frame by frame, it took on average 8.1 ± 3.8 minutes to manually implement the counting method and input the values into pre-constructed templates for analysis. Taking into account all the process from frame extraction to the final score obtained per video, the time is therefore about 10 minutes for a rater manually counting and often more for high bubble scores.

From both field experiments presented in this section, it is clear that the VGE evolution pattern post dive varies even for the same exposure. Previous studies relying on grading from Doppler recordings confirm these findings. Looking at unrestricted recreational dives monitored only once 20-40min post dive between 1989 and 1991 [351] showed that the likelihood of finding a high bubble grade (HBG) decreased by a factor of 0.78 for each 10min delay in recording compared to their planned timeline of recording (20-40min post dive). The authors concluded that delaying monitoring by 25-30min could underestimate VGE peak by 20%. They also found a positive correlation between likelihood of Doppler HBG with exposure severity, repetitive diving, male and/or older divers, but a lower incidence of HBG over the course of the days from multi-day diving trips pointing to possible adaptation. More recently, a review paper looked at Doppler and B-mode VGE monitoring protocols and concluded that the current protocols, most often monitoring twice post-dive around the 30 and 90 min points, are not frequent enough [134]. In a study with frequent post-dive measurements every 15 and 30min using Doppler KM grading [352], the authors found very different results than other studies regarding the influence of pre-dive exercise, which could also be linked to their more frequent measurement protocol.

The next step regarding this work is an obvious one: correlating all the physiological measurements and evolutions post-dive, in particular with regards to possible new emboli detection techniques attempted (Prof. Karapantsios’ team on the one hand, and our

178

Ultrasound Pulse Inversion technique on the other). A recent paper using Dual Frequency Ultrasound on swine showed that small emboli in tissue could be detected [353] and that the evolution of VGE and small “tissue” emboli differed significantly in some animals. Another new literature addition partly supports the hypothesis of this work regarding FMD post-dive evolution to bubble size: it was shown that agitated saline injection had a measurable effect on FMD-measured vascular dysfunction in PFO positive subjects only [354].

In conclusion, having the capability to record echocardiograms at multiple timepoints post- dive, in conjunction with other physiological parameters presents advantages. Nevertheless, since each echocardiogram takes around 10 minutes to analyse manually, it is clearly important to go towards automatisation of the counting methodology for maximum use and benefit of this new assessment technique.

179

5.3 Automatizing the counting of VGE on echocardiograms12

5.3.1 Introduction

Decompression algorithms should of course aim to prevent DCS, but because of the low incidence of DCS and for ethics reasons, currently the only usable measure for their evaluation is the number of Venous Gas Emboli (VGE) present after the dive. After diving, decompression bubbles are often detectable as VGE circulating in the right heart cavities as these are normally "filtered out" by the lungs, as shown in Figure 55. There is a statistical association between higher VGE numbers and risk for DCS. A quantitative evaluation of VGE is thus an important tool, but until now, no automatic quantification was reported.

The aim of this section is to build upon the counting methodology validated in Section 5.1 and semi-automatize it in MatLab. Indeed, the method presented in Section 5.1 was shown to be better at discriminating intermediate VGE levels; offers a linear continuous scale of assessment; and is more consistent without expert training but time-consuming. Use of the method in Section 5.2 during collaborative field experiments showed that raters needed around 10 minutes per video to manually implement the method. As multiple timepoints post dive are clearly advantageous, this ends up being a significant time investment from experimenters, with our own field experiment yielding over 130 echocardiograms over a 2 day period. As such, the possibility to semi-automate the method with the aim to decrease the time spent on data analysis, whilst keeping a similarly high level of repeatability as per the human rater validation, is advantageous.

As per the methodology validated, here bubbles are also counted to assess decompression stress: frames are selected during the heart cycle for which the tricuspid valves are fully open, then the counts from ten consecutive heart cycles are averaged. Finally, the results of the automated counting are compared to the agreement between human raters.

12 The automatization of VGE counting was the subject of an MSc project by Joe How Hui which I supervised directly. Please refer to the acknowledgements section. 180

Figure 55: Example frame of post-dive echocardiography showing VGE in the right heart chambers (right) and corresponding anatomy of the heart (left).

5.3.2 Methods

5.3.2.1 Data and statistical comparison

The same set of ultrasound B-mode sequences which was used to validate the counting methodology for repeatability and agreement between raters is again used in this section. The sequences were acquired from 10 volunteers post dive and both without and with leg flexion: modified 4-chamber view B-mode echocardiography for rest and stress conditions.

Our results from the MatLab program are compared to the known agreement between human raters (n=9) assessed on the same set of 20 videos presented in 5.1. The agreement for human raters was shown to give an intra-class correlation coefficient (ICC) of 0.84 (95% CI: from 0.77 to 0.92). The goal here with the semi-automatized method is therefore to demonstrate that it is possible to get close to (or above) this ICC agreement.

A time-comparison is also performed looking at the user-input time needed to use the final semi-automated method and comparing this to the known time for manually counting the bubbles as per the methodology in Sections 5.1 and 5.2. The computer time needed to process the videos is also quoted, although this is something that can be further improved in the future and not directly limiting in terms of experimenter data analysis time.

181

5.3.2.2 Brief overview of the image processing steps

Here we present first the overall strategy used in semi-automatizing the manual counting method. Additional details are presented thereafter (Section 5.3.2.3 for the detailed processing and then Section 5.3.3.2 for threshold setting). The overall strategy for automatizing the manual counting method is shown in Figure 56 and Figure 57 which lists the different steps implemented in semi-automatizing the method.

Figure 56 presents one frame from the video data at different steps of the processing. The cone-shaped field of view is extracted after prompting the user to select the 3 apexes of the cone. A binary threshold is then applied and the septum identified by detecting the intensity boundary from each horizontal line of the image. A series of morphological opening and closing operations allow identifying the right heart chamber ROI. Finally, bubbles are identified as regions bigger than a certain area and of higher intensity.

Figure 56: (a) Example frame ; (b) Septum identification from intensity boundary; (c) segmentation ROI result; (d) Identified bubbles within the ROI

182

Figure 57: Overview of the image processing steps: a user-set intensity threshold (T) is used to segment the right heart chambers defined as the ROI; the area of the ROI evolution in time is then used to identify frames where the heart valves are open (maximum area) for bubble counting.

Following from Figure 57, the reasoning behind steps A and B are given below:

Step A: ROI segmentation of the right heart chambers

The right heart chamber area is segmented and the area size used to identify the frames for which the heart valves are open; the assumption being here that the maximum heart area will correspond to the opening of the valves. This was checked in practice and works well, especially since the variation of bubbles on consecutive frames is small: as long as the valves are not in the middle of the ROI (they could be miscounted as bubbles depending on intensity thresholding in this case), then it does not matter which frame exactly is chosen from those to count.

Step B: Selection of open valves frames for bubble counting

The area of the ROI is also used to identify heart cycles as it varies with the beating of the heart and stretching of the muscle. In practice when the valves close then the ROI becomes either the right ventricle or right atrium, then when the tricuspid valves open the ROI is both of these together, significantly increasing the area. The frames selected for counting are in the

183 end of the diastolic phase / proto-systolic phase of the heartbeat, with the tricuspid valves that separate the right atrium and ventricle open.

Once the correct frames in each heart cycle is identified, the bubbles are counted on each frame (from an intensity and size threshold), as well as the frame immediately prior and immediately consecutive to the identified one. The average is calculated from these 3 frames and taken as the number of bubbles for this heart cycle. The procedure is repeated over 10 heart cycles for each video.

5.3.2.3 Additional processing steps details

Segmentation of the right heart chambers (step A)

The user selects one representative frame from the video and the three cone vertices from the ultrasound imaging field of view, then performs a rough right heart chamber segmentation and bubble clicking. These manual inputs are used to semi-automatize the initialisation parameters for the different videos, as discussed in the results section in more details. The intensity threshold found from user input is used to binarise the image, and any small unconnected object outside the right heart chamber regions is discarded (noise removal).

A connected component analysis is then used (MatLab function “bwlabel”) to identify the largest vertical connected region as the septum. Similarly, if the heart muscle in the left lower corner is not connected already to the septum then the two groups are linked by extending a straight line between their two closest boundary points (MatLab function “intline”), as shown in Figure 58. The rest of the cone field outline determines the left upper boundaries of the ROI. The area of the ROI is extracted for each frame of the video and stored for further analysis.

184

Figure 58: Use of the connected component labelling for identification of the septum and muscle wall. In the case where the binarisation operation (example result shown in a) disconnects the two structures (figure b) these are linked by a straight line (shown in c and d) to avoid counting structures in the arterial side and outside the cone shaped field of view.

Selection of open valves frames (step B)

From the segmentation of the right heart chambers in step A, the ROI area is plotted with respect to the frame index of the video (time). The area is indeed expected to vary periodically with each heartbeat. Outlier removal is also performed prior to plotting by setting a maximum allowable difference between the areas of two consecutive frames. The set variation is set by using the mean for that particular video.

The graphical scatter distribution is then fitted with a spline smoothing model, and the ten first peaks detected are stored as the frames to count bubbles in (open tricuspid valves).

Figure 59 confirms the hypothesis that the segmented area against the frame index of the video shows an oscillatory pattern as the heart beats. After outlier removal, the data points are fitted with a spline smoothing model, shown in Figure 60, and peaks are identified. Image

185 segmentation therefore enables automated frame selection where the frames to retain are the ones corresponding to the selected peaks and their immediate prior and consecutive frames.

Figure 59: Plot of ROI area against frame number

Figure 60: Corresponding sinusoidal fitting to Figure 59

Bubble counting

For each identified frame index in step B, the previous and next frames are also analysed and the average of the three taken as the value for that frame. This is then repeated for all 10 open valve frames, so in total each video yields 30 frames on which bubbles need to be counted.

186

The heart mask (ROI segmentation) created from step A is used to only keep the inside of the heart chamber where bubbles appear, with minimal heart muscle boundary.

A connected component analysis is performed after binary intensity threshold with a parameter T2. This parameter T2 is initialised using the clicked bubbles from the user input at the beginning of the processing (semi-automatization). The user input is also used to initialise a size window for bubbles from the clicked bubbles that are identified by the operator originally. For each connected component in the ROI post T2 binarisation (potential bubble), a size criterion is therefore applied, getting rid of any remaining heart muscle.

Finally, the averaged counting over 10 cycles is presented as the result for the video to the operator, together with the identified bubbles on the corresponding frames, Figure 62.

Figure 61: Bubble counting steps shown for one frame a/ after threshold T for the whole cone region; b/ corresponding identified ROI; c/ Masking of steps a and b for bubble counting

Figure 62: Corresponding original frame and final output with counted bubbles to Figure 61

187

5.3.3 Results

5.3.3.1 Variation assessment and agreement with manual raters

Using the semi-automated counting method on the 20 videos, an idea of the variation of counted bubbles is given below, where the mean for these was found at 12.3 bubbles:

 Standard error associated to averaging the bubble counts on the frame identified and the ones before and after that one: 0.75  Standard deviation of the bubbles counted over 10 cardiac cycles: 2.8

This shows that the exact correct frame index identification will not significantly influence the results of bubble number.

The calculated intra-class correlation coefficient between the bubbles counted on each of the videos by the algorithm and the gold standard (as defined in Section 5.1) is found to be r=0.87 (95% CI of: 0.66 to 1). This is comparable to the variation found between human raters, of 0.84 (95% CI: from 0.77 to 0.92), showing that the method can be relatively well semi-automatized. However trying more videos, initialised with the semi-automated method discussed above, for instance from Section 5.2 where data was collected on different experiments, is definitely needed to validate this in a more general setting.

5.3.3.2 Time spent

For the segmentation a non-supervised method was preferred as it does not require a large training set manually annotated by the operator. A bottom-up approach encompassing edge detection and morphological operations [355, 356] is thus used to allow a low computational time. This geometry and intensity based method is dependent on correct threshold setting for each video but is significantly faster than supervised methods [357]. This bottom-up approach could even in theory be implemented “real-time” on GPU for continuous post-dive monitoring.

With the intensity threshold already set manually for each video, it takes on average about 3 minutes to count the bubbles on each video using the bottom-up approach. Due to time constraints it was not possible to re-run a full analysis with different segmentation methods such as active-contour snakes or registration from original manual segmentation but these 188 methods were tried on individual frames and take upwards of 30 seconds per frame so they would be significantly more time-consuming.

5.3.3.3 Semi-automatizing of the thresholds used

The methods for image processing selected here are straight-forward and rely on prior anatomical knowledge given the specific conditions of these videos. They are heavily dependent on the appropriate intensity threshold value to binaries the ROI, which differs from video to video. This section aims to present initial results on how to semi-automatize the threshold used in the bottom-up approach presented here (parameter T in Figure 57). A possible approach assumes a Gaussian Mixture Model intensity distribution [358]. The videos used in this section (same as 5.1) are in trichromatic green, red and blue components (not grayscale), Figure 64. The trichromatic green coefficient is found to present a Gaussian Mixture Model intensity distribution so this is used thereafter for attempting to semi- automatize the threshold initial value.

Figure 63: Original frame and corresponding trichromatic coefficients

The user is asked to roughly select the area of the right heart chamber including a section of the heart muscle on the first frame of the video, as illustrated in Figure 64.

189

Figure 64: Example of user segmentation used for threshold intensity calculation

Figure 65 shows the intensity histogram from this segmented area. A moving average is used to fit the trend, before setting the intensity threshold.

Figure 65: Histogram of the trichromatic green component, showing a characteristic bimodal distribution

190

Figure 66: Example of 2 term Gaussian Mixture Model curve-fitted histogram for the trichromatic green component

A comparison with manual segmentation of ROI on one video shows correct pattern identification even with automatic T setting. Since this is used only for choosing which frames to count bubbles in (and the absolute area is not of interest) the systematic error with semi-automatizing the threshold initial parameter does not create additional problems.

Figure 67: Comparison of the area of the ROI with respect to the frame index for one video. In blue: manual user segmentation and in red: semi-automatic segmentation area.

191

5.3.4 Discussion

It was shown that an algorithm relying on some initial input from the users can be automatized to reproduce the counting methodology for VGE on echocardiograms. The automatic method could be used post-dive to assess how number of bubbles during first 3 hours post dive changes by taking regular measurements. This could resolve the current issue of missing the peak bubble number with current monitoring methods, something also highlighted from the field experiments in Section 5.2. The decay curve shape might provide useful information with regards to characterising decompression stress of the dive profile.

Our method avoids misclassification of valves as bubbles by selecting frames where valves are fully open. This is particularly useful in allowing relatively easy automatizing the counting method computationally. However correct frame identification is dependent on segmentation quality. This in turn is significantly easier when the right heart chamber muscles remain fully visible in the whole video and movement of the probe on the subject’s skin is minimised during acquisition. Interestingly the flex (stress) echocardiograms often had worse videos in terms of (often out of plane) motion of the heart. It is not surprisingly crucial that the best possible data acquisition is done in the first place: correct view selection and probe placement allows significantly better performance of the algorithm and also of manual raters. Some new probe guidance positioning systems specifically for B-mode echocardiography do exist [359, 360] and these would go a long way in assuring best quality for non-cardiologist echographers in field experiments.

The variation of bubbles counted per frame was also more important for videos recorded during flexion, as the bubbles seem to flow in bigger numbers during the flexion resulting in a pulsatile bubble number fluctuation corresponding to each flexion with a delay. Looking for systematic bias in the number of bubbles counted in all identified frame indexes (from a Bland-Altman plot, similarly to those in Section 5.1) could be considered to apply a correction factor in the future if needed.

192

Segmentation Area (pixel)

Frame index

Figure 68: Heart rhythm and ROI variation illustrated for a flex (stress) echocardiogram showing a region (black circle) with higher variation than for typical rest recordings.

Finally, it should be noted that the methods for counting in the right heart chambers can be extended to also count any bubbles appearing in the left heart chambers (arterial side). Although not commonly observed, there are instances where bubbles in the arterial circulation are seen, even when the pulmonary circulation is not overloaded. The pulmonary gas exchange gets rid of any bubble bigger than about 8 μm in theory. The presence of arterio-venous shunts in the heart or lungs, however, offers a paradoxical entry point for bubbles into the arterial blood and left heart chambers (please refer to Section 2.3 for details about this filtering).

The strategy implemented in processing the echocardiograms semi-automatically is linked to the aim of achieving a short processing time. In particular, we aim to limit the time spent by the human rater compared to him manually counting the whole video. The bottom up approach takes 3min per video to run without user input selection. Adding the time for a user to flick through some frames of the video and perform the segmentation and bubble identification required on one frame brings this to 5min. It should however be noted that if the segmentation is done using a different method this time is significantly increased (over 10min for registration from user input or snake implementation) in favour of robustness

193 against initialisation parameters, so the future work will need to consider what the best compromise is.

5.3.3 Conclusion

A semi-automated method was developed in MatLab to identify the septum and use it to segment the right chambers of the heart. The frames for which the valves are open are then successfully identified and bubbles are counted in these. An initial comparison of the results with the manual method shows good agreement with the assessment from expert raters, as demonstrated by an intra-class correlation coefficient of r=0.87 (95% CI: 0.66 to 1) and compared to the human-human agreement of 0.84 (95% CI: 0.77 to 0.92). This suggests that this method could potentially become a powerful tool for quantitative evaluation of the validity of decompression algorithms.

Taking the work forward, the full method with automatic initialisation on all videos should be tested to compare agreement. A particularly important point relates also to the quality of the original echocardiography recordings: having the heart visible throughout the video, correct probe positioning, etc. In this respect and for use in field experiments by non-cardiologists, coupling this methodology with an automated probe positioning guidance system would be extremely useful.

194

Chapter 6 – Conclusion

This chapter provides a summary of the contributions reported in the thesis and discusses these in the context of the aims defined in Chapter 2. The impact of the work with respect to the research field is assessed and limitations are identified. Proposed future directions are also presented.

6.1 Summary of contributions

The research reported in the thesis was driven by the aim to develop experimental and computational methods going towards the quantification of a hyperbaric stress index. A detailed literature review was undertaken looking at bubble formation and circulatory bubble dynamics in the context of human hyperbaric exposures to propose the aims of the thesis.

The first goal of the project, in Chapter 3, was to implement a decompression simulation and analysis platform in MatLab and in particular analyse real dive profiles from the DAN Europe database with respect to the Bühlmann ZHL16 algorithm used by diving computer manufacturers. The platform was created and validated for calculating staged ascents from various depths, then modified to input real dive profiles and track the partial pressures of compartments throughout the dive for further analyses. From a first analysis on real profiles from the database, and for the range of profiles considered, we found as expected that these existing models are poor predictors of accidents and demonstrate that ascent rate seems to be an important predictor of DCS.

The second aim, in Chapter 4, tackled the fact that the precise formation site and growth mechanism of decompression bubbles in vivo remains unknown. A novel experimental set-up and analysis code for the real-time optical study of decompression induced bubble growth dynamics was developed. This set-up was used to study bubble growth from nitrogen saturated water on ex-vivo rabbit muscle and fat tissues. We showed that the role of the substrate from which bubble detach plays a significant role. Bubble density, nucleation threshold, detachment size and coalescence behaviour were shown to be significantly different for the two substrates, whereas growth rates after a critical size are governed by diffusion as expected, and a competition for dissolved gas between adjacent multiple bubbles

195 was demonstrated. The role of the tissue substrate, decoupled from the absorption of gas with this new experimental set-up, was therefore demonstrated to also play a role in bubble growth, merging/non-merging behaviour, as well as detachment from decompression. These findings are not accounted for in current modelling efforts so our experimental set-up could be used in the development of a more physiologically relevant decompression model.

The third question, in Chapter 5, of the thesis looked at optimising the circulatory bubble evaluation endpoint post dive. Vascular circulating bubbles are normally assessed semi- quantitatively by trained human raters who grade the severity on echocardiograms. A new counting methodology was compared to the current assessment and found to perform significantly better in terms of repeatability between raters (external) and for the same rater (internal). This was also achieved with very minimal training by non-experts. The counting method offers a continuous scale but is time-consuming to implement. Therefore, the method seems well suited for use in interventional human diving experiments, where it is ethically impossible to subject volunteer divers to dive profiles generating extreme bubble grades. It was used in two collaborative field experiments, taking around 10 minutes per video. These experiments demonstrated the importance of regular post-dive measurement time points. Moreover the method is suitable for the development of automated counting software. Image processing techniques to semi-automate this method were implemented. Segmentation of the right chambers of the heart is used to identify the frames for which the valves are open and bubbles are counted in these. An initial comparison of the results with the manual method showed good agreement with the assessment from expert raters: intra-class correlation coefficient between the bubbles counted on each of the videos by the algorithm and the gold standard is found to be r=0.87 (95% CI of: 0.66 to 1). However this is dependent on the correct threshold value for masking as well as the quality of the recorded echocardiograph.

The proposed techniques could be used towards optimising the cardiovascular risk assessment of hyperbaric decompression stress caused by circulatory bubble dynamics. As is discussed in the literature review, one component of this new evaluation endpoint for decompression modelling has to include post dive circulating bubbles in the bloodstream (VGE), as well as other physiological markers. The advantage of going beyond DCS outcome is that preconditioning interventions and other inter-personal physiological differences could then be assessed, aiming at limiting the environmental stress instead of just preventing symptoms. With new evidence of long term effects of even recreational scuba diving with no

196 history of DCS, and ethics considerations in experimenting on humans, this is an important consideration.

6.2 Discussion of limitations and future work

This work sets the foundations for developing new methods towards quantifying hyperbaric exposure induced cardiovascular stress and in particular decompression emboli. At the same time it opens up a number of questions that should be tackled in the future.

With respect to Chapter 3 dealing with a computational decompression platform implementation for analysing real dive profiles, the lack of physiological data in current databases significantly limits the potential conclusions. Therefore having access to a bigger database with more real diving profiles, ideally matching this with physiological parameters, would be advantageous in evaluating the ascent rate conclusions. Suggestions for additional safety from our findings would be to lower the permitted ascent rate to around 9-10 msw/min if further analysis with more profiles confirms the results. In addition, it would be interesting to re-analyse all the profiles with respect to free gas phase tracking algorithms and assess whether some of these could be a better predictor of accidents in the range of profiles available. The main difficulty in this respect, however, relates to the lack of published parameter values used in initialising the only commercially implemented free gas phase model, which would severely limit any conclusions applicable to dive computers.

In Chapter 4 a new experimental set-up was developed for optically studying bubble growth and density on ex-vivo tissue substrates during decompression in real-time. Although the experiments reported treated the case of air saturated distilled water and ex vivo rabbit muscle and fat tissues from 3 to 0 bar decompression at 1 bar/min, this can be extended to other liquid or gas properties and composition (eg blood substitute for pH and viscosity effects), temperatures, decompression profile and nucleation site (tissue surface structure and hydrophobicity). A design modification of the set-up could be considered for a follow-up of this experiment to include flow and looking at live endothelial cell cultures instead of ex-vivo substrates is the obvious next step. The set-up size could also allow for animal experimentation, where physiological effects would be better studied.

197

For Chapter 5 where the evaluation endpoint for circulating bubbles on echocardiography post dive was assessed, a limitation of the counting methodology proposed is the time it takes for the raters to implement this manually. In assessing the potential for semi-automatizing the counting method via image-processing, the current agreement with human raters was found promising for manually-set intensity thresholds, but a full evaluation with the automated setting needs to be performed to take the work forward. Another key factor relates also to the quality of the original echocardiography recordings, since having the heart visible throughout the video and correct probe positioning is crucial. This is particularly difficult in field experiments typically performed in less than ideal conditions on portable machines. In this respect and for use in field experiments by non-cardiologists, coupling this methodology with an automated probe positioning guidance system would be extremely useful.

6.3 Final thoughts

The field of human hyperbaric exposure study, in particular with the aim to prevent decompression sickness, has been around for a long time. However moving from just symptom prevention to reduction in stress is a new approach that relates to other extreme environmental exposures.

The methods and results presented in this thesis constitute a small addition to extending our capabilities for decompression induced human cardiovascular stress. It should be clear from the discussion above that this is a non-trivial problem to explore physiologically and that many factors contribute to inter-personal differences documented in the literature. Considerable efforts are still needed to devise controlled physiological experiments and quantify these effects, and investing into developing novel methods is a part of that endeavour.

198

Appendix A - PUBLICATIONS

JOURNAL PAPERS

Papadopoulou V, Eckersley RJ, Balestra C, Karapantsios TD, Tang M-X. A critical review of physiological bubble formation in hyperbaric decompression. Advances in colloid and interface science. 2013;191–192(0):22-30.

Theunissen S, Guerrero F, Sponsiello N, Cialoni D, Pieri M, Germonpré P, Obeid G, Tillmans F, Papadopoulou V, Hemelryck W, Marroni A, De Bels D, Balestra C. Nitric oxide- related endothelial changes in breath-hold and scuba divers. Undersea Hyperb Med. 2013 Mar-Apr;40(2):135-44.

Hemelryck W, Germonpré P, Papadopoulou V, Rozloznik M, Balestra C. Long term effects of recreational SCUBA diving on higher cognitive function. Scand J Med Sci Sports. 2014; 24:928-934.

Germonpré P, Papadopoulou V, Hemelryck W, Obeid G, Eckersley RJ, Tang M-X, Balestra C. The use of portable 2D echocardiography and "frame-based" bubble counting as a tool to evaluate diving decompression stress. Diving and Hyperbaric Medicine; 2014 Mar;44(1):5- 13.

Papadopoulou V, Tang M-X, Balestra C, Eckersley RJ, Karapantsios TD. Circulatory Bubble Dynamics: From Physical to Biological Aspects. (Invited) Advances in colloid and interface science. 2014; 206:239-249.

Buzzacott P, Lambrechts K, Mazur A, Wang Q, Papadopoulou V, Theron M, Balestra C, Guerrero F. A ternary model of decompression sickness in rats. Computers in Medicine and Biology. 2014; 55:74-78.

Buzzacott P, Schuster A, Gerges A, Hemelryck W, Lambrechts K, Madden D, Papadopoulou V, Tkachenko Y, Mazur A, Tillmans F, Rozloznik M, Wang Q, Mollerlokken A, François G, Arne S. A New Model of Head-Up Display Dive Computer Addressing Safety-Critical Rate of Ascent and Returning Gas Pressure - A Pilot Trial. International Journal of Computer Science in Sport. 2014; 13:50-58.

199

Buzzacott P, Papadopoulou V, Baddeley A, Petri NM, Lind F. Theoretical tissue compartment inert gas pressures during a deep dive with and without deep decompression stops: a case analysis. International Journal of Maritime Health. 2015; 66(1):36-42.

Cavalade M, Papadopoulou V, Theunissen S, Balestra C. Heart Rate Variability and Critical Flicker Fusion Frequency changes during and after free fall in experienced skydivers. European Journal of Applied Physiology. 2015; 2015 Feb 26. [Epub ahead of print] PMID: 25715913.

Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Kostoglou M, Tang MX, Karapantsios TD. Decompression induced bubble dynamics on ex-vivo fat and muscle tissue surfaces with a new experimental set up. Colloids and Surfaces B: Biointerfaces. 2015; 129:121-129.

CONFERENCE PRESENTATIONS

First Author

Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Tang M-X, Karapantsios, T. A study of decompression induced bubble dynamics on different tissue surfaces with a novel experimental set-up. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

Papadopoulou V, Evgenidis S, Eckersley RJ, Mesimeris T, Balestra C, Tang M-X, Karapantsios, T. Decompression induced bubble growth on tissue surfaces. Smart and Green Interfaces Conference - COST MP1106, Marseille, 22-24 April 2014.

Papadopoulou V, Balestra C. Effect of preconditioning on SCUBA diving decompression stress. Smart and Green Interfaces Conference - COST MP1106, Marseille, 22-24 April 2014.

Papadopoulou V, Evgenidis S, Eckersley RJ, Balestra C, Tang M-X, Karapantsios, T. Decompression induced bubble growth on tissue surfaces from gas saturated solutions. Tri- continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013.

200

Papadopoulou V, Hui J, Balestra C, Hemelryck W, Germonpré P, Eckersley R, Tang M. Evaluating the counting of venous gas emboli on post-scuba dive echocardiographs. Tri- continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013.

Papadopoulou V, Hui J, Balestra C, Hemelryck W, Germonpré P, Eckersley R, Tang M. Automated Counting of Venous Gas Emboli in Post-SCUBA Dive Echocardiography. IEEE- UFFC conference, Prague, Czech Republic, 21-25 July 2013.

Papadopoulou V, Evgenidis S, Eckersley RJ, Balestra C, Tang M-X, Karapantsios, T. Effect of different tissue surfaces on decompression induced bubble growth from gas saturated solutions. COST MP1106 conference, Prague, Czech Republic, 20-24 March 2013.

Papadopoulou V, Eckersley RJ, Balestra C, Tang M-X. Decompression modelling: a projection procedure for dissolved phase tracking simulations. EUBS 2012 conference, Belgrade, Serbia, 11-15 Sept 2012.

Other

Ward M, Yildiz Y, Papadopoulou V, Eckersley RJ, Tang M-X. Modelling of ultrasound contrast agent oscillations in vessels based on an infinite mirror image method. 2015 IEEE International Ultrasonics Symposium, Taipei, Taiwan, 21-24 Oct 2015.

Evgenidis SP, Zacharias K, Papadopoulou V, Theunissen S, Balestra C, Karapantsios TD. Post-dive detection of bubbles in scuba divers employing electrical impedance spectroscopy measurements. EUBS2015 conference, Amsterdam, 19-23 Sept 2015.

Hemelryck W, Calistri J, Papadopoulou V, Theunissen S, Schmitz M, Balestra C. The use of echography to assess cervical spine impact on first row rugby players: an innovative measure for injury prevention . The 2015 Sports Science Summit, London, UK, 13-15 Jan 2015.

Popov G, Balestra C, Lafère P, Papadopoulou V, Rozloznik M, Tillmans F, Hemelryck W, Distefano G. Investigating the cognitive function of trained free-divers with a psychometric test in the mares icon net ready dive computer – a feasibility study. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

201

Lafère P, Germonpré P, HemelryckW, Papadopoulou V, Balestra C. Early-stage brain activation during hyperbaric exposure is oxygen-related. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

Rozloznik M, Stebelova K, Papadopoulou V, Tillmans F, Tkachenko Y, Hemelryck W, Zeman M, Pudil R, Balestra C. Coping with an underwater life-threatening event during SCUBA diving: a case study. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

Buzzacott P, Schuster A, Gerges A, Hemelryck W, Lambrechts K, Madden D, Papadopoulou V, Tkachenko Y, Mazur A, Tillmans F, Rozloznik M, Wang Q, Mollerlokken A, Guerrero F, Sieber A. A new model of head up display dive computer addressing safety-critical rate of ascent and returning gas pressure: a pilot trial. EUBS2014 conference, Wiesbaden, 14-27 Sept 2014.

Balestra C, Theunissen S, Lafère P, Papadopoulou V, Rozloznik M, Hemelryck W, Germonpré P. Preconditioning interventions to reduce decompression stress in scuba divers. 6th International Congress of Medicine in Space and Extreme Environments (ICMS), Berlin, Germany, 16-19 Sept 2014.

Rozloznik M, Tillmans F, Stebelova K, Papadopoulou V, Tkachenko Y, Hemelryck W, Zeman M, Pudil R, Balestra C. TRIMIX Deep Dive Research 2013. International Workshop of Evidence-Based Hyperbaric Medicine, 4th Ostrava’s Days of Hyperbaric Medicine, Rožnov pod Radhoštěm, Czech Republic, 18-20 June 2014.

Balestra C, Theunissen S, Rozloznik M, Tillmans F, Hemelryck W, Papadopoulou V, Cialoni D, Pieri M, Marroni A, Germonpré P. Patency of the cardiac foramen ovale (PFO) and circulating gas emboli; is there a problem? X. World Congress on High Altitude Medicineand Physiology & Mountain Emergency Medicine, Bolzano, Italy, 25-31 May 2014.

Balestra C, De Bels D, Rozloznik M, Tillmans F, Hemelryck W, Papadopoulou V, Cialoni D, Pieri M, Marroni A, Germonpré P. “The normobaric oxygen paradox” Triggering EPO without hypoxia. X. World Congress on High Altitude Medicineand Physiology & Mountain Emergency Medicine, Bolzano, Italy, 25-31 May 2014.

Buzzacott P, Papadopoulou V, Baddeley A, Petri N, Lind F. Exceptional diving exposure with deep stops and DCS (case study). Tri-continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013. 202

Hemelryck W, Balestra C, Papadopoulou V, Rozloznik M, Germonpré P, Lafère P. Effects of normobaric oxygen breathing on human brain executive functions. Tri-continental scientific meeting on diving and hyperbaric medicine, Reunion Island, 23-28 Sept 2013.

Rozloznik M, Hemelryck W, Tillmans F, Papadopoulou V, Theunissen S, Balestra C. Effect of Pre-Dive Hydration on Venous Gas Bubble Production in Divers. Hyperbaric and Aerospace Medicine Congress, Prague, Czech Republic, 23-24 May 2013.

Outreach

Papadopoulou V, Decompression induced bubble growth (the physics), "New Perspectives in Diving Medicine" 10th International DAN Divers Day, Wiesbaden, Germany, 28 Sept 2014.

Papadopoulou V, Lambrechts K, Balestra C, Guerrero F. Interest of physiology of decompression (PHYPODE) research for application to space environments. UK Space Environments Conference, National Space Centre, Leicester, UK, 9-10 Nov 2013.

Balestra C, Papadopoulou V, Distefano G, Popov G, Guerrero F. The Phypode Project, new advances in Physiopathology of Diving. COST MP1106 conference, Prague, Czech Republic, 20-24 March 2013.

BOOK CHAPTERS

Balestra C, Cialoni D, Buzzacott P, Hemelryck W, Papadopoulou V, Pieri M, Marroni A. Chapter: Recreational diving today: decompression habits, DAN Europe database insights. In: Balestra C, Germonpré P, The Science of Diving: Things your instructor never told you. 2014: 13-36; Lambert Academic Publishing, ISBN: 978-3-659-66233-.

Papadopoulou V, Chatterton J, Popov G, Eckersley JE, Balestra C, Karapantsios TD, Tang MX, Cialoni D, Kot J. Chapter: Decompression Theory. In: Balestra C, Germonpré P, The Science of Diving: Things your instructor never told you. 2014: 13-36; Lambert Academic Publishing, ISBN: 978-3-659-66233-1.

203

204

Appendix B - CV

205

206

207

208

References

[1] V. Papadopoulou, R.J. Eckersley, C. Balestra, T.D. Karapantsios, M.-X. Tang, A critical review of physiological bubble formation in hyperbaric decompression, Adv Colloid Interface Sci, 191–192 (2013) 22-30. [2] P. Germonpre, V. Papadopoulou, W. Hemelryck, G. Obeid, P. Lafere, R.J. Eckersley, M.X. Tang, C. Balestra, The use of portable 2D echocardiography and 'frame-based' bubble counting as a tool to evaluate diving decompression stress, Diving Hyperb Med, 44 (2014) 5-13. [3] V. Papadopoulou, M.X. Tang, C. Balestra, R.J. Eckersley, T.D. Karapantsios, Circulatory bubble dynamics: From physical to biological aspects, Adv Colloid Interfac, 206 (2014) 239-249. [4] V. Papadopoulou, S. Evgenidis, R.J. Eckersley, T. Mesimeris, C. Balestra, M. Kostoglou, M.X. Tang, T.D. Karapantsios, Decompression induced bubble dynamics on ex vivo fat and muscle tissue surfaces with a new experimental set up, Colloids and surfaces. B, Biointerfaces, 129 (2015) 121-129. [5] C. Acott, A brief history of diving and decompression illness, South Pacific Underwater Medicine Society Journal, 29 (1999) 98-109. [6] R. Boyle, New pneumatical experiments about respiration, Philos Trans, 5 (1670) 2011-2058. [7] R. Jackson, Rails Across the Mississippi: A History of the St Louis Bridge, University of Illinois Press2001. [8] D. McCullough, The Great Bridge: The Epic Story of the Building of the Brooklynn Bridge, Simon and Shuster1983. [9] P. Bert, Barometric pressure (1878). Translation by Hitchcock MA and Hitchcock FA. Columbus , Ohio: College Book Company, 1943 Republished by Undersea and Hyperbaric Medical Society, Bethesda, 1978. [10] Mixed Gas Diving: the Ultimate Challenge for Technical Diving, Watersport Publishing1993. [11] R.Y. Nishi, A.O. Brubakk, O.S. Eftedal, Bubble Detection, in: A.O. Brubakk, T.S. Neuman (Eds.) Bennett and Elliott’s Physiology and Medicine of Diving, WB Saunders, Philadelphia, 2003, pp. 501- 529. [12] J. Hugon, L. Barthelemy, J.C. Rostain, B. Gardette, The pathway to drive decompression microbubbles from the tissues to the blood and the lymphatic system as a part of this transfer, Undersea Hyperb Med, 36 (2009) 223-236. [13] C. Balestra, The lymphatic pathway for microbubbles, Diving Hyperb Med, 44 (2014) 1. [14] T.W. Beck, S. Daniels, W.D.M. Paton, E.B. Smith, Detection of bubbles in decompression sickness, Nature, 276 (1978) 173-174. [15] M.P. Spencer, Decompression limits for compressed air determined by ultrasonically detected blood bubbles, J Appl Physiol, 40 (1976) 229-235. [16] R.E. Reinertsen, V. Flook, S. Koteng, A.O. Brubakk, Effect of oxygen tension and rate of pressure reduction during decompression on central gas bubbles, J Appl Physiol, 84 (1998) 351-356. [17] M. Ljubkovic, Z. Dujic, A. Mollerlokken, D. Bakovic, A. Obad, T. Breskovic, A.O. Brubakk, Venous and arterial bubbles at rest after no-decompression air dives, Medicine and science in sports and exercise, 43 (2011) 990-995. [18] M.P. Spencer, S.D. Campbell, J.L. Sealey, F.C. Henry, J. Lindbergh, Experiments on decompression bubbles in the circulation using ultrasonic and electromagnetic flowmeters, Journal of occupational medicine. : official publication of the Industrial Medical Association, 11 (1969) 238- 244. [19] M.P. Spencer, D.C. Johanson, Investigation of New Principles for Human Decompression Schedules Using Doppler Ultrasonic Blood Bubble Detection, Technical Report to ONR on contract N00014-73-C-0094, Institute for Environmental Medicine and Physiology, Seattle, 1974. [20] J. Imbert, D. Paris, J. Hugon, The Arterial Bubble Model for Decompression Tables Calculations, EUBS conference, 2004 [Available at:

209 http://gtuem.praesentiertihnen.de/tools/literaturdb/project2/pdf/Imbert%20JP.%20- %20EUBS%202004.pdf]. [21] R.D. Vann, J.J. Freiberger, J.L. Caruso, Divers Alert Network report on decompression illness, diving fatalities and project dive exploration: 2005 edition (based on 2003 data), DAN technical Report, 2005. [22] PADI, Professional Association of Diving Instructors (PADI) Worldwide certification history from 1967 to 2006. Available at http://www.padi.com/padi/en/footerlinks/certhistorynum.aspx (accessed February 11, 2012). [23] R.D. Vann, D.M. Uguccioni, DAN's annual review of recreational scuba diving injuries and fatalities based on 1998 data., DAN technical report 2000. [24] N.W. Pollock, Divers Alert Network Annual diving report 2008, in: N.W. Pollock (Ed.) DAN technical report, 2008. [25] R.D. Vann, Mechanisms and Risks of Decompression, in: A.A. Bove (Ed.) Bove and Davis' Diving Medicine, WB Saunders, Philadelphia, 2004, pp. 127-164. [26] P.D. Cooper, C. Van den Broek, D.R. Smart, Hyperbaric chamber attendant safety II: 14-year staff health review of multiplace chamber attendants, Diving Hyperb Med, 39 (2009) 71-76. [27] D.J. Doolette, S.J. Goble, C.J. Pirone, Health outcome of hyperbaric chamber inside attendants following compressed-air exposure and oxygen decompression, SPUMS J, 34 (2004) 63-67. [28] G. Ladd, V. Stepan, L. Stevens, The Abacus project: establishing the risk of recreational scuba death and decompression illness, SPUMS J, 32 (2002) 124-128. [29] N.W. Pollock, Divers Alert Network Annual diving report 2007 (based on 2005 data), DAN technical report, Divers Alert Network, Durham, 2007. [30] T.E. Berghage, D. Durman, US Navy air decompression schedule risk analysis, Technical Report, Naval Medical Research and Development Command, 1980. [31] D.J. Temple, R. Ball, P.K. Weathersby, E.C. Parker, S. Survanshi, The dive profile and manifestations of decompression sickness cases after air and nitrogen-oxygen dives. volume i: data set summaries, manifestation descriptions, and key files, US Navy technical report, Naval Medical Research Centre (NMRC), 1999. [32] H.D. Van Liew, E.T. Flynn, Decompression tables and dive-outcome data: graphical analysis, Undersea Hyperb Med, 32 (2005) 187-198. [33] A. Fahlman, D.M. Dromsky, Dehydration effects on the risk of severe decompression sickness in a swine model, Aviat Space Environ Med, 77 (2006) 102-106. [34] E. Gempp, J.E. Blatteau, Preconditioning methods and mechanisms for preventing the risk of decompression sickness in scuba divers: a review, Res Sports Med, 18 (2010) 205-218. [35] A.A. Bove, Risk of decompression sickness with patent foramen ovale, Undersea Hyperb Med, 25 (1998) 175-178. [36] G. Jayalalitha, R. Uthayakumar, Fractal approach to understand PFO and DCS in sport divers, Fractals, 18 (2010) 499-511. [37] V. Lisignoli, A.M. Lanzone, D. Zavalloni, P. Pagnotta, P. Presbitero, Closure of patent foramen ovale: when and how?, Current vascular pharmacology, 5 (2007) 322-327. [38] C.T. Leffler, Effect of ambient temperature on the risk of decompression sickness in surface decompression divers, Aviat Space Environ Med, 72 (2001) 477-483. [39] C.B. Toner, R. Ball, The effect of temperature on decompression and decompression sickness risk: A critical review, US Navy technical report, Naval Medical Research Center (NMRC), 2004. [40] J.P. Dervay, M.R. Powell, B. Butler, C.E. Fife, The effect of exercise and rest duration on the generation of venous gas bubbles at altitude, Aviat Space Environ Med, 73 (2002) 22-27. [41] J.R. Claybaugh, Y.-C. Lin, Exercise and decompression sickness: a matter of intensity and timing, The Journal of Physiology, 555 (2004) 588. [42] U. Wisloff, A.O. Brubakk, Aerobic endurance training reduces bubble formation and increases survival in rats exposed to hyperbaric pressure, The Journal of physiology, 537 (2001) 607-611.

210

[43] L.W. Jankowski, R.Y. Nishi, D.J. Eaton, A.P. Griffin, Exercise during decompression reduces the amount of venous gas emboli, Undersea Hyperb Med, 24 (1997) 59-65. [44] U. Wisloff, R.S. Richardson, A.O. Brubakk, Exercise and nitric oxide prevent bubble formation: a novel approach to the prevention of decompression sickness?, The Journal of physiology, 555 (2004) 825-829. [45] P.K. Weathersby, L.D. Homer, E.T. Flynn, On the likelihood of decompression sickness, J Appl Physiol, 57 (1984) 815-825. [46] A. Boussuges, P. Blanc, F. Molenat, E. Bergmann, J.M. Sainty, Haemoconcentration in neurological decompression illness, Int J Sports Med, 17 (1996) 351-355. [47] D.Z.H. Levett, I.L. Millar, Bubble trouble: a review of diving physiology and disease, Postgraduate Medical Journal, 84 (2008) 571-578. [48] NAVSEA, U.S. Navy Diving Manual, 6 ed., US Navy. Available at: http://www.supsalv.org/00c3_publications.asp2008. [49] DAN, Underwater Diving Accident including Oxygen First Aid Manual, Divers Alert Network, Durham, 1992. [50] J. Pennefather, C. Edmonds, C. Lowry, R. Walker, Diving and Subaquatic Medicine, 4 ed., E Arnold, London, 2002. [51] P. Cianci, J.B. Slade, Jr., Delayed treatment of decompression sickness with short, no-air-break tables: review of 140 cases, Aviat Space Environ Med, 77 (2006) 1003-1008. [52] R.E. Moon, P.J. Sheffield, Guidelines for treatment of decompression illness, Aviat Space Environ Med, 68 (1997) 234-243. [53] E. Gempp, J.E. Blatteau, Risk factors and treatment outcome in scuba divers with spinal cord decompression sickness, Journal of critical care, 25 (2010) 236-242. [54] L.K. Weaver, Monoplace hyperbaric chamber use of U.S. Navy Table 6: a 20-year experience, Undersea Hyperb Med, 33 (2006) 85-88. [55] E.D. Thalmann, Principles of US Navy recompression treatments for decompression sickness, in: R.E. Moon, P.J. Sheffield (Eds.) 45th Undersea and Hyperbaric Medical Society Workshop, Undersea and Hyperbaric Medical Society, Kensington, 1996, pp. 75-95. [56] Proceedings of Validation of Dive Computers Workshop, in: S.L. Blogg, M.A. Lang, A. Møllerløkken (Eds.), Akademika Publishing, Gdansk, Poland, 24 August 2011. [57] A. Sieber, M. Stoianova, E. Jöbstl, E. Azzopardi, M.D.J. Sayer, M.F. Wagner, Dive Computers: The Need for Validation and Standards, in: S.L. Blogg, M.A. Lang, A. Møllerløkken (Eds.) Validation of Dive Computers Workshop, Akademika Publishing, Gdansk, Poland, 24 August 2011, pp. 29-43. [58] G.C.C. Daman, A.E. Boycott, J.S. Haldane, The prevention of compressed air illness, Journal of Hygiene, 8 (1908) 342-443. [59] R.D. Workman, Calculations of decompression schedules for nitrogen-oxygen and helium- oxygen dives, Technical report, Research Report 6-65. US NEDU, 1965. [60] A.A. Bühlmann, Decompression-Decompression Sickness, Berlin New York: Springer-Verlag1984. [61] A.A. Buhlmann, [Decompression problems in diving in mountain lakes], Schweizerische Zeitschrift fur Sportmedizin, 37 (1989) 80-83; discussion 99-102. [62] R.J. Harris, D.J. Doolette, D.C. Wilkinson, D.J. Williams, Measurement of fatigue following 18 msw dry chamber dives breathing air or enriched air nitrox, Undersea Hyperb Med, 30 (2003) 285- 291. [63] R.L. Pyle, The Importance of Deep Safety Stops: Rethinking Ascent Patterns From Decompression Dives. , South Pacific Underwater Medical Society Journal (SPUMS) 27 (1997) 112- 115. [64] B.A. Hills, A thermodynamic and kinetic approach to decompression sickness. PhD Thesis, The University of Adelaide, 1966. [65] D.E. Yount, Skins of varying permeability: A stabilization mechanism for gas cavitation nuclei, The Journal of the Acoustical Society of America, 65 (1979) 1429-1439.

211

[66] D.E. Yount, Application of a bubble formation model to decompression sickness in rats and humans, Aviat Space Environ Med, 50 (1979) 44-50. [67] R.Y. Nishi, L.A. Kuehn, Digital computation of decompression profiles, Technical report, DCIEM, Report 884, 1973. [68] R.Y. Nishi, G.R. Lauckner, Development of the dciem 1983 decompression model for compressed air diving, Technical report, DCIEM Report 44-R-44, 1984. [69] R.Y. Nishi, Decompression tables for compressed air diving based on kidd-stubbs 1971 model, Technical report, DCIEM Report 82-R-42, 1982. [70] J.E. Blatteau, E. Gempp, C. Balestra, T. Mets, P. Germonpre, Predive sauna and venous gas bubbles upon decompression from 400 kPa, Aviat Space Environ Med, 79 (2008) 1100-1105. [71] P. Germonpre, J.M. Pontier, E. Gempp, J.E. Blatteau, S. Deneweth, P. Lafere, A. Marroni, C. Balestra, Pre-dive vibration effect on bubble formation after a 30-m dive requiring a decompression stop, Aviat Space Environ Med, 80 (2009) 1044-1048. [72] T.G. Leighton, The Acoustic Bubble, Academic Press, London, 1994. [73] J. Campbell, The tribonucleation of bubbles, Journal of Physics D: Applied Physics, 1 (1968) 1085. [74] A.T.J. Hayward, Tribonucleation of bubbles, British Journal of Applied Physics, 18 (1967) 641. [75] K.G. Ikels, Production of gas bubbles in fluids by tribonucleation, J Appl Physiol, 28 (1970) 524- 527. [76] P.K. Weathersby, L.D. Homer, E.T. Flynn, Homogeneous nucleation of gas bubbles in vivo, J Appl Physiol, 53 (1982) 940-946. [77] E.N. Harvey, D.K. Barnes, W.D. McElroy, A.H. Whiteley, D.C. Pease, K.W. Cooper, Bubble formation in animals. I. Physical factors, Journal of Cellular and Comparative Physiology, 24 (1944) 1- 22. [78] J.-E. Blatteau, J.-B. Souraud, E. Gempp, A. Boussuges, Gas Nuclei, Their Origin, and Their Role in Bubble Formation, Aviation, Space, and Environmental Medicine, 77 (2006) 1068-1076. [79] L. Hill, Caisson sickness and the physiology of work in compressed air, E. Arnold, London, 1912. [80] T.R. Hennessy, H.V. Hempleman, An examination of the critical released gas volume concept in decompression sickness, Proc R Soc Lond B Biol Sci, 197 (1977) 299-313. [81] F.E. Fox, K.F. Herzfeld, Gas Bubbles with Organic Skin as Cavitation Nuclei, The Journal of the Acoustical Society of America, 26 (1954) 984-989. [82] M. Greenspan, C.E. Tschiegg, Radiation-induced acoustic cavitation; apparatus and some results, Journal of Research of the National Bureau of Standards. Section C: Engineering and Instrumentation, 71C (1967) 299-315. [83] M.S. Plesset, S.S. Sadhal, On the stability of gas bubbles in liquid-gas solutions, Applied Scientific Research, 38 (1982) 133-141. [84] D.E. Yount, R.H. Strauss, Bubble formation in gelatin: A model for decompression sickness, Journal of Applied Physics, 47 (1976) 5081-5089. [85] D.E. Yount, On the evolution, generation, and regeneration of gas cavitation nuclei, The Journal of the Acoustical Society of America, 71 (1982) 1473-1481. [86] L.D. Landau, E.M. Lifshitz, Statistical Physics, 3 ed., Butterworth-Heinemann, Oxford, 1980. [87] D.E. Yount, T.D. Kunkle, Gas nucleation in the vicinity of solid hydrophobic spheres, Journal of Applied Physics, 46 (1975) 4484-4486. [88] D.N. Walder, Adaptation to decompression sickness in caisson work in: S.W. Tromp, W.H. Weihe (Eds.) Third International Biometeorology Congress, Pergamon, St Pau, France, 1967, pp. 350-359. [89] M.P. Spencer, H.F. Clarke, Precordial monitoring of pulmonary gas embolism and decompression bubbles, Aerospace medicine, 43 (1972) 762-767. [90] A. Evans, D.N. Walder, Significance of Gas Micronuclei in the Aetiology of Decompression Sickness, Nature, 222 (1969) 251-252. [91] R.D. Vann, J. Grimstad, C.H. Nielsen, Evidence for gas nuclei in decompressed rats, Undersea Biomed Res, 7 (1980) 107-112. 212

[92] P.S. Epstein, M.S. Plesset, On the Stability of Gas Bubbles in Liquid-Gas Solutions, The Journal of Chemical Physics, 18 (1950) 1505-1509. [93] B.R. Wienke, Reduced gradient bubble model, International journal of bio-medical computing, 26 (1990) 237-256. [94] B.R. Wienke, Numerical phase algorithm for decompression computers and application, Computers in biology and medicine, 22 (1992) 389-406. [95] N. Gaskins, R.D. Vann, E. Hobbs, M. Swingle, S. Lee, D. Needham, Surface tension and bubble formation in agar gelatin, Undersea Hyperb Med, 28 (2001) 56-59. [96] M. Strasberg, The Onset of Ultrasonic Cavitation in Tap Water, Catholic University of America, 1956. [97] R.H. Strauss, T.D. Kunkle, Isobaric bubble growth: a consequence of altering atmospheric gas, Science, 186 (1974) 443-444. [98] F.F. Abraham, Homogeneous nucleation theory; the pretransition theory of vapor condensation. Advances in theoretical chemistry; supplement 1, Academic Press, New York, 1974. [99] S. Goldman, The stability of bubbles formed from supersaturated solutions, and homogeneous nucleation of gas bubbles from solution, both revisited, J Phys Chem B, 112 (2008) 16701-16709. [100] N. Ishida, T. Inoue, M. Miyahara, K. Higashitani, Nano Bubbles on a Hydrophobic Surface in Water Observed by Tapping-Mode Atomic Force Microscopy, Langmuir, 16 (2000) 6377-6380. [101] N. Ishida, M. Sakamoto, M. Miyahara, K. Higashitani, Attraction between Hydrophobic Surfaces with and without Gas Phase, Langmuir, 16 (2000) 5681-5687. [102] E.E. Meyer, Q. Lin, J.N. Israelachvili, Effects of dissolved gas on the hydrophobic attraction between surfactant-coated surfaces, Langmuir, 21 (2005) 256-259. [103] J.W.G. Tyrrell, P. Attard, Images of Nanobubbles on Hydrophobic Surfaces and Their Interactions, Phys Rev Lett, 87 (2001) 176104. [104] S. Yang, S.M. Dammer, N. Bremond, H.J.W. Zandvliet, E.S. Kooij, D. Lohse, Characterization of Nanobubbles on Hydrophobic Surfaces in Water, Langmuir, 23 (2007) 7072-7077. [105] R. Arieli, A. Marmur, Decompression sickness bubbles: Are gas micronuclei formed on a flat hydrophobic surface?, Resp Physiol Neurobi, 177 (2011) 19-23. [106] B.M. Borkent, S.M. Dammer, H. Schonherr, G.J. Vancso, D. Lohse, Superstability of surface nanobubbles, Phys Rev Lett, 98 (2007) 204502. [107] S. Goldman, Free energy wells for small gas bubbles in soft deformable materials, The Journal of Chemical Physics, 132 (2010) 164509-164513. [108] S. Goldman, Generalizations of the Young-Laplace equation for the pressure of a mechanically stable gas bubble in a soft elastic material, J Chem Phys, 131 (2009) 184502. [109] M. Chappell, S. Payne, A crevice bubble growth model for the analysis of decompression sickness, Conf Proc IEEE Eng Med Biol Soc, 3 (2005) 2240-2243. [110] M.A. Chappell, S.J. Payne, A physiological model of gas pockets in crevices and their behavior under compression, Respir Physiol Neurobiol, 152 (2006) 100-114. [111] M.A. Chappell, S.J. Payne, A physiological model of the release of gas bubbles from crevices under decompression, Respir Physiol Neurobiol, 153 (2006) 166-180. [112] M.E. Burkard, H.D. Van Liew, Simulation of exchanges of multiple gases in bubbles in the body, Respir Physiol, 95 (1994) 131-145. [113] P.P. Foster, B.D. Butler, Decompression to altitude: assumptions, experimental evidence, and future directions, J Appl Physiol, 106 (2009) 678-690. [114] M.A. Chappell, S.J. Payne, The effect of cavity geometry on the nucleation of bubbles from cavities, J Acoust Soc Am, 121 (2007) 853-862. [115] N. Divinis, T.D. Karapantsios, M. Kostoglou, C.S. Panoutsos, V. Bontozoglou, A.C. Michels, M.C. Sneep, R. De Bruijn, H. Lotz, Bubbles growing in supersaturated solutions at reduced gravity, Aiche J, 50 (2004) 2369-2382. [116] N. Divinis, T.D. Karapantsios, R. de Bruijn, M. Kostoglou, V. Bontozoglou, J.C. Legros, Bubble dynamics during degassing of liquids at microgravity conditions, Aiche J, 52 (2006) 3029-3040. 213

[117] N. Divinis, M. Kostoglou, T.D. Karapantsios, V. Bontozoglou, Self-similar growth of a gas bubble induced by localized heating: the effect of temperature-dependent transport properties, Chem Eng Sci, 60 (2005) 1673-1683. [118] N. Divinis, T.D. Karapantsios, R. de Bruyn, M. Kostoglou, V. Bontozoglou, J.C. Legros, Lateral motion and interaction of gas bubbles growing over spherical and plate heaters, Microgravity Sci Tec, 18 (2006) 204-209. [119] T.D. Karapantsios, M. Kostoglou, N. Divinis, V. Bontozoglou, Nucleation, growth and detachment of neighboring bubbles over miniature heaters, Chem Eng Sci, 63 (2008) 3438-3448. [120] T.D. Karapantsios, M. Kostoglou, S.P. Evgenidis, From single bubbles on solid surfaces to massive bubbly flows during decompression sickness, In: Proceedings of the Symposium "Life in Space for Life on Earth"Angers, France, 2008. [121] S.P. Evgenidis, N.A. Kazakis, T.D. Karapantsios, Bubbly flow characteristics during decompression sickness: Effect of surfactant and electrolyte on bubble size distribution, Colloids Surf A Physicochem Eng Asp, 365 (2010) 46-51. [122] M.A. Chappell, S.J. Payne, A physiological model of the interaction between tissue bubbles and the formation of blood-borne bubbles under decompression, Physics in Medicine and Biology, 51 (2006) 2321. [123] H.D. Van Liew, M.E. Burkard, Density of decompression bubbles and competition for gas among bubbles, tissue, and blood, J Appl Physiol, 75 (1993) 2293-2301. [124] E.D. Thalmann, E.C. Parker, S.S. Survanshi, P.K. Weathersby, Improved probabilistic decompression model risk predictions using linear-exponential kinetics, Undersea Hyperb Med, 24 (1997) 255-274. [125] M.A. Chappell, S. Uzel, S.J. Payne, Modeling the detachment and transport of bubbles from nucleation sites in small vessels, IEEE transactions on bio-medical engineering, 54 (2007) 2106-2108. [126] C.R. Gutvik, A.O. Brubakk, A Dynamic Two-Phase Model for Vascular Bubble Formation During Decompression of Divers, Biomedical Engineering, IEEE Transactions on, 56 (2009) 884-889. [127] C.R. Gutvik, R.G. Dunford, Z. Dujic, A.O. Brubakk, Parameter estimation of the copernicus decompression model with venous gas emboli in human divers, Medical & biological engineering & computing, 48 (2010) 625-636. [128] E.N. Harvey, Physical factors in bubble formation, in: J.F. Fulton (Ed.) Decompression Sickness, Saunders, London, 1951, pp. 90-114. [129] S.A. Mohammadein, K.G. Mohamed, Concentration distribution around a growing gas bubble in tissue, Mathematical biosciences, 225 (2010) 11-17. [130] R.S. Srinivasan, W.A. Gerth, M.R. Powell, Mathematical model of diffusion-limited gas bubble dynamics in unstirred tissue, J Appl Physiol, 86 (1999) 732-741. [131] J. Hugon, J.C. Rostain, B. Gardette, A new biophysical decompression model for estimating the risk of articular bends during and after decompression, Journal of theoretical biology, 283 (2011) 168-179. [132] B.R. Wienke, Reduced Gradient Bubble Model in Depth, Best Publishing Company, Florida, 2003. [133] L.E. Howle, P.W. Weber, R.D. Vann, M.C. Campbell, Marginal DCS events: their relation to decompression and use in DCS models, J Appl Physiol, 107 (2009) 1539-1547. [134] S.L. Blogg, M. Gennser, The need for optimisation of post-dive ultrasound monitoring to properly evaluate the evolution of venous gas emboli, Diving Hyperb Med, 41 (2011) 139-146. [135] V.K. Kumar, R.D. Billica, J.M. Waligora, Utility of Doppler-detectable microbubbles in the diagnosis and treatment of decompression sickness, Aviat Space Environ Med, 68 (1997) 151-158. [136] I.B. Parlak, S.M. Egi, A. Ademoglu, C. Balestra, P. Germonpre, A. Marroni, S. Aydin, A Neuro- fuzzy Approach of Bubble Recognition in Cardiac Video Processing Digital Information and Communication Technology and Its Applications, in: H. Cherifi, J.M. Zain, E. El-Qawasmeh (Eds.) Digital Information and Communication Technology and Its Applications; Communications in Computer and Information Science, Springer, Berlin Heidelberg, 2011, pp. 277-286. 214

[137] K. Tufan, A. Ademoglu, E. Kurtaran, G. Yildiz, S. Aydin, S.M. Egi, Automatic detection of bubbles in the subclavian vein using Doppler ultrasound signals, Aviat Space Environ Med, 77 (2006) 957-962. [138] I.B. Parlak, S.M. Egi, A. Ademoglu, C. Balestra, P. Germonpre, A. Marroni, Intelligent bubble recognition on cardiac videos using Gabor wavelet, International Journal of Digital Information and Wireless Communications, 1 (2011) 195-203 [139] J.G. Swan, B.D. Bollinger, T.G. Donoghue, J.C. Wilbur, S.D. Phillips, D.L. Alvarenga, D.A. Knaus, P.J. Magari, J.C. Buckey, Microbubble detection following hyperbaric chamber dives using dual- frequency ultrasound, J Appl Physiol, 111 (2011) 1323-1328. [140] J.C. Buckey, D.A. Knaus, D.L. Alvarenga, M.A. Kenton, P.J. Magari, Dual-frequency ultrasound for detecting and sizing bubbles, Acta Astronautica, 56 (2005) 1041-1047. [141] S.J. Payne, M.A. Chappell, Automated determination of bubble grades from Doppler ultrasound recordings, Aviat Space Environ Med, 76 (2005) 771-777. [142] M.A. Chappell, S.J. Payne, A method for the automated detection of venous gas bubbles in humans using empirical mode decomposition, Ann Biomed Eng, 33 (2005) 1411-1421. [143] P.P. Foster, A.H. Feiveson, R. Glowinski, M. Izygon, A.M. Boriek, A model for influence of exercise on formation and growth of tissue bubbles during altitude decompression, American Journal of Physiology - Regulatory, Integrative and Comparative Physiology, 279 (2000) R2304- R2316. [144] A. Mollerlokken, T. Breskovic, I. Palada, Z. Valic, Z. Dujic, A.O. Brubakk, Observation of increased venous gas emboli after wet dives compared to dry dives, Diving Hyperb Med, 41 (2011) 124-128. [145] V. Flook, The physics and physiology of decompression, Eur J Underw Hyperb Med, 1 (2000) 8– 13. [146] W.W. Mapleson, An electrical analogue for uptake and exchange of inert gases and other agents, J Appl Physiol 18 (1963) 197–204. [147] CNES, Space Physiology, Cepadues, Toulouse, 1983. [148] C. Balestra, A. Marroni, B. Farkas, P. Peetrons, F. Vanderschueren, E. Duboc, T. Snoeck, P. Germonpré, The fractal approach as a tool to understand asymptomatic brain hyperintense MRI signals, Fractals, 12 (2004) 67-72. [149] A. Arefmanesh, S.G. Advani, E.E. Michaelides, An Accurate Numerical-Solution for Mass Diffusion-Induced Bubble-Growth in Viscous-Liquids Containing Limited Dissolved-Gas, International Journal of Heat and Mass Transfer, 35 (1992) 1711-1722. [150] P. Payvar, Mass Transfer-Controlled Bubble-Growth during Rapid Decompression of a Liquid, International Journal of Heat and Mass Transfer, 30 (1987) 699-706. [151] H.J. Yoo, C.D. Han, Oscillatory Behavior of a Gas Bubble Growing (or Collapsing) in Viscoelastic Liquids, Aiche J, 28 (1982) 1002-1009. [152] J.K. Edzwald, J.P. Malley, C. Yu, A Conceptual-Model for Dissolved Air Flotation in Water- Treatment, Water Supply, Vol 9, No 1, (1991) 141-150. [153] C.M. Muth, E.S. Shank, Gas embolism, The New England journal of medicine, 342 (2000) 476- 482. [154] A. Suzuki, D.M. Eckmann, Embolism bubble adhesion force in excised perfused microvessels, Anesthesiology, 99 (2003) 400-408. [155] M.P. Poland, A.J. Sutton, T.M. Gerlach, Magma degassing triggered by static decompression at Kilauea Volcano, Hawai'i, Geophys Res Lett, 36 (2009). [156] N.G. Lensky, O. Navon, V. Lyakhovsky, Bubble growth during decompression of magma: experimental and theoretical investigation, J Volcanol Geoth Res, 129 (2004) 7-22. [157] J.E. Gardner, M. Hilton, M.R. Carroll, Bubble growth in highly viscous silicate melts during continuous decompression from high pressure, Geochim Cosmochim Ac, 64 (2000) 1473-1483. [158] P. Zehner, R. Benfer, Modelling fluid dynamics in multiphase reactors, Chemical Engineering Science, 51 (1996) 1735-1744.

215

[159] K. Heide, E. Hartmann, T. Stelzner, R. Muller, Degassing of a cordierite glass melt during nucleation and crystallization, Thermochim Acta, 280 (1996) 243-250. [160] V. Papadopoulou, R.J. Eckersley, C. Balestra, T.D. Karapantsios, M.X. Tang, A critical review of physiological bubble formation in hyperbaric decompression, Adv Colloid Interface Sci, 191-192 (2013) 22-30. [161] B.A. Hills, D.C. Grulke, Evaluation of ultrasonic bubble detectors in vitro using calibrated microbubbles at selected velocities, Ultrasonics, 13 (1975) 181-184. [162] J. Lubbers, J.W. Van Den Berg, An ultrasonic detector for microgasemboli in a bloodflow line, Ultrasound in medicine & biology, 2 (1977) 301-310. [163] H.D. Van Liew, J. Conkin, M.E. Burkard, How big are decompression-sickness bubbles?, Proceedings of the XIXth Annual Meeting of European Undersea Biomedical Society, SINTEF Unimed, Trondheim, Norway, 1993, pp. 258-262. [164] B.A. Hills, B.D. Butler, Size distribution of intravascular air emboli produced by decompression, Undersea biomedical research, 8 (1981) 163-170. [165] J. Lee, S. Kentish, M. Ashokkumar, Effect of surfactants on the rate of growth of an air bubble by rectified diffusion, The journal of physical chemistry. B, 109 (2005) 14595-14598. [166] W. Pugsley, L. Klinger, C. Paschalis, T. Treasure, M. Harrison, S. Newman, The Impact of Microemboli during Cardiopulmonary Bypass on Neuropsychological Functioning, Stroke, 25 (1994) 1393-1399. [167] Y. Tian, J.A. Ketterling, R.E. Apfel, Direct observation of microbubble oscillations, The Journal of the Acoustical Society of America, 100 (1996) 3976-3978. [168] D. Cosgrove, Ultrasound contrast agents: an overview, European journal of radiology, 60 (2006) 324-330. [169] T.D. Mast, Empirical relationships between acoustic parameters in human soft tissues, Acoustics Research Letters Online, 1 (2000) 37-42. [170] M.J. Stewart, Contrast echocardiography, Heart, 89 (2003) 342-348. [171] M.X. Tang, H. Mulvana, T. Gauthier, A.K. Lim, D.O. Cosgrove, R.J. Eckersley, E. Stride, Quantitative contrast-enhanced ultrasound imaging: a review of sources of variability, Interface focus, 1 (2011) 520-539. [172] J. Wu, J. Tong, Measurements of the nonlinearity parameter B/A of contrast agents, Ultrasound Med Biol, 24 (1998) 153-159. [173] R.J. Browning, H. Mulvana, M.X. Tang, J.V. Hajnal, D.J. Wells, R.J. Eckersley, Effect of albumin and dextrose concentration on ultrasound and microbubble mediated gene transfection in vivo, Ultrasound Med Biol, 38 (2012) 1067-1077. [174] S. Hernot, A.L. Klibanov, Microbubbles in ultrasound-triggered drug and gene delivery, Advanced drug delivery reviews, 60 (2008) 1153-1166. [175] J. Rooze, E.V. Rebrov, J.C. Schouten, J.T. Keurentjes, Dissolved gas and ultrasonic cavitation--a review, Ultrasonics sonochemistry, 20 (2013) 1-11. [176] P. Johansen, Mechanical heart valve cavitation, Expert Rev Med Devic, 1 (2004) 95-104. [177] P. Johansen, B.R. Travis, P.K. Paulsen, H. Nygaard, J.M. Hasenkam, Cavitation caused by mechanical heart valve prostheses - A review, Apmis, 111 (2003) 108-112. [178] J. Bessereau, N. Genotelle, C. Chabbaut, A. Huon, A. Tabah, J. Aboab, S. Chevret, D. Annane, Long-term outcome of iatrogenic gas embolism, Intensive care medicine, 36 (2010) 1180-1187. [179] R. Engelman, The neurologic complications of cardiac surgery: introduction, Seminars in thoracic and cardiovascular surgery, 13 (2001) 147-148. [180] R.A. Baker, M.J. Andrew, J.L. Knight, Evaluation of neurologic assessment and outcomes in cardiac surgical patients, Seminars in thoracic and cardiovascular surgery, 13 (2001) 149-157. [181] J.W. Hammon, D.A. Stump, J.B. Butterworth, D.M. Moody, Approaches to reduce neurologic complications during cardiac surgery, Seminars in thoracic and cardiovascular surgery, 13 (2001) 184-191.

216

[182] P.B. Gorelick, Stroke prevention therapy beyond antithrombotics: unifying mechanisms in ischemic stroke pathogenesis and implications for therapy: an invited review, Stroke, 33 (2002) 862- 875. [183] T. Brott, J. Bogousslavsky, Drug therapy - Treatment of acute ischemic stroke, New Engl J Med, 343 (2000) 710-722. [184] B.D. Butler, B.A. Hills, The lung as a filter for microbubbles, J Appl Physiol Respir Environ Exerc Physiol, 47 (1979) 537-543. [185] P. Germonpre, P. Dendale, P. Unger, C. Balestra, Patent foramen ovale and decompression sickness in sports divers, J Appl Physiol, 84 (1998) 1622-1626. [186] D. Madden, M. Lozo, Z. Dujic, M. Ljubkovic, Exercise after SCUBA diving increases the incidence of arterial gas embolism, J Appl Physiol, (2013). [187] G.J. Tortora, B.H. Derrickson, Principles of Anatomy & Physiology, John Wiley & Sons, Inc.2012, pp. 816. [188] M.P. Spencer, Y. Oyama, Pulmonary capacity for dissipation of venous gas emboli, Aerospace Med., 42 (1971) 822-827. [189] T.J.R. Francis, S.J. Mitchell, Chap.10.4 Pathophysiology of decompression sickness, in: A.O. Brubakk, T.S. Neuman (Eds.) Bennett and Elliott's Physiology and Medicine of Diving, Saunders2003, pp. 530-556. [190] C.J. Harvey, M.J. Blomley, R.J. Eckersley, R.A. Heckemann, J. Butler-Barnes, D.O. Cosgrove, Pulse-inversion mode imaging of liver specific microbubbles: improved detection of subcentimetre metastases, Lancet, 355 (2000) 807-808. [191] M. Kostoglou, T.D. Karapantsios, Bubble dynamics during the non-isothermal degassing of liquids. Exploiting microgravity conditions, Advances in colloid and interface science, 134-35 (2007) 125-137. [192] S.F. Jones, G.M. Evans, K.P. Galvin, Bubble nucleation from gas cavities - a review, Adv Colloid Interface Sci, 80 (1999) 27-50. [193] B.B. Mikic, W.M. Rohsenow, P. Griffith, On bubble growth rates, International Journal of Heat and Mass Transfer, 13 (1970) 657-666. [194] M.S. Plesset, S.A. Zwick, The Growth of Vapor Bubbles in Superheated Liquids, Journal of Applied Physics, 25 (1954) 493-500. [195] L.E. Scriven, On the dynamics of phase growth: L. E. Scriven, Chem. Engng. Sci. 10: 1–13, 1959, Chem Eng Sci, 50 (1995) 3905. [196] L. Rayleigh, Pressure due to collapse of bubbles, Philosophical Magazine, 94 (1917). [197] L.T. Thuy, R.H. Weiland, Mechanisms of Gas Desorption from Aqueous Solution, Industrial & Engineering Chemistry Fundamentals, 15 (1976) 286-293. [198] R.H. Weiland, L.T. Thuy, A.N. Liveris, Transition from Bubbling to Quiescent Desorption of Dissolved-Gases, Industrial & Engineering Chemistry Fundamentals, 16 (1977) 332-335. [199] S.D. Lubetkin, Why is it much easier to nucleate gas bubbles than theory predicts?, Langmuir, 19 (2003) 2575-2587. [200] P.H. Schweitzer, V.G. Szebehely, Gas Evolution in Liquids and Cavitation, Journal of Applied Physics, 21 (1950) 1218-1224. [201] G. Burrows, F. Preece, The process gas evolution from low vaporpressure liquids upon reduction of pressure, Trans. Inst. Chem. Eng, 32 (1954) 84-89. [202] W.E. Langlois, Similarity rules for isothermal bubble growth, J Fluid Mech, 15 (1963) 111-118. [203] M. Barak, Y. Katz, Microbubbles: pathophysiology and clinical implications, Chest, 128 (2005) 2918-2932. [204] P.B. Duncan, D. Needham, Test of the Epstein-Plesset model for gas microparticle dissolution in aqueous media: effect of surface tension and gas undersaturation in solution, Langmuir, 20 (2004) 2567-2578. [205] H. Kierzkowska-Pawlak, A. Chacuk, Pressure Swing Absorption of Carbon Dioxide in Dmepeg Solutions, Environ Prot Eng, 35 (2009) 37-45. 217

[206] H. Hikita, Y. Konishi, Desorption of Carbon-Dioxide from Supersaturated Water in an Agitated Vessel, Aiche J, 30 (1984) 945-951. [207] M.A. Chappell, S.J. Payne, The effect of cavity geometry on the nucleation of bubbles from cavities, J Acoust Soc Am, 121 (2007) 853-862. [208] M.A. Chappell, S.J. Payne, A physiological model of the release of gas bubbles from crevices under decompression, Resp Physiol Neurobi, 153 (2006) 166-180. [209] C.F. Burrows, K.C. Bovee, Metabolic changes due to experimentally induced rupture of the canine urinary bladder, American journal of veterinary research, 35 (1974) 1083-1088. [210] C. Colin, J. Fabre, J. McQuillen, Bubble and slug flow at microgravity conditions: State of knowledge and open questions, Chem Eng Commun, 141 (1996) 155-173. [211] G. Duhar, C. Colin, Dynamics of bubble growth and detachment in a viscous shear flow, Physics of Fluids, 18 (2006). [212] G. Duhar, C. Colin, A predictive model for the detachment of bubbles injected in a viscous shear flow with small inertial effects, Physics of Fluids, 16 (2004) L31-L34. [213] J.F. Klausner, R. Mei, D.M. Bernhard, L.Z. Zeng, Vapor Bubble Departure in Forced-Convection Boiling, International Journal of Heat and Mass Transfer, 36 (1993) 651-662. [214] R. Arieli, A. Marmur, Evolution of bubbles from gas micronuclei formed on the luminal aspect of ovine large blood vessels, Resp Physiol Neurobi, 188 (2013) 49-55. [215] M.A. Chappell, S. Uzel, S.J. Payne, Modeling the Detachment and Transport of Bubbles From Nucleation Sites in Small Vessels, Biomedical Engineering, IEEE Transactions on, 54 (2007) 2106- 2108. [216] T. Lango, T. Morland, A.O. Brubakk, Diffusion coefficients and solubility coefficients for gases in biological fluids and tissues: a review, Undersea Hyperb Med, 23 (1996) 247-272. [217] A. Kabalnov, D. Klein, T. Pelura, E. Schutt, J. Weers, Dissolution of multicomponent microbubbles in the bloodstream: 1. Theory, Ultrasound Med Biol, 24 (1998) 739-749. [218] R.S. Srinivasan, W.A. Gerth, M.R. Powell, A mathematical model of diffusion-limited gas bubble dynamics in tissue with varying diffusion region thickness, Resp Physiol, 123 (2000) 153-164. [219] R.S. Srinivasan, W.A. Gerth, M.R. Powell, Mathematical models of diffusion-limited gas bubble dynamics in tissue, J Appl Physiol, 86 (1999) 732-741. [220] H.D. Van Liew, J. Conkin, M.E. Burkard, The oxygen window and decompression bubbles: estimates and significance, Aviat Space Environ Med, 64 (1993) 859-865. [221] H.D. Van Liew, M.E. Burkard, Bubbles in circulating blood: stabilization and simulations of cyclic changes of size and content, J Appl Physiol, 79 (1995) 1379-1385. [222] M.P. Hlastala, H.D. Van Liew, Absorption of in vivo inert gas bubbles, Respir Physiol, 24 (1975) 147-158. [223] F. Dexter, B.J. Hindman, Recommendations for hyperbaric oxygen therapy of cerebral air embolism based on a mathematical model of bubble absorption, Anesthesia and analgesia, 84 (1997) 1203-1207. [224] A.B. Branger, D.M. Eckmann, Theoretical and experimental intravascular gas embolism absorption dynamics, J Appl Physiol, 87 (1999) 1287-1295. [225] D.M. Eckmann, A.B. Branger, D.P. Cavanagh, Gas embolism, The New England journal of medicine, 342 (2000) 2000-2001; author reply 2001-2002. [226] A.B. Branger, C.J. Lambertsen, D.M. Eckmann, Cerebral gas embolism absorption during hyperbaric therapy: theory, J Appl Physiol, 90 (2001) 593-600. [227] A.B. Branger, D.M. Eckmann, Accelerated arteriolar gas embolism reabsorption by an exogenous surfactant, Anesthesiology, 96 (2002) 971-979. [228] D.M. Eckmann, S.C. Armstead, Surfactant Reduction of Cerebral Infarct Size and Behavioral Deficit in a Rat Model of Cerebrovascular Arterial Gas Embolism, Journal of Applied Physiology, (2013). [229] L. Rayleigh, VIII. On the pressure developed in a liquid during the collapse of a spherical cavity, Philosophical Magazine Series 6, 34 (1917) 94-98. 218

[230] M.S. Plesset, The dynamics of cavitation bubbles, Journal of Applied Mechanics, Transactions ASME, 16 (1949) 228. [231] D.B. Khismatullin, Gas microbubbles and their use in medicine, in: A.A. Doinikov (Ed.) Bubble and Particle Dynamics in Acoustic Fields: Modern Trends and Applications, Research Signpost, Kerala 2005, pp. 231–289. [232] D.B. Khismatullin, A. Nadim, Radial oscillations of encapsulated microbubbles in viscoelastic liquids, Physics of Fluids, 14 (2002) 3534-3557. [233] A. Prosperetti, The Thermal-Behavior of Oscillating Gas-Bubbles, J Fluid Mech, 222 (1991) 587- 616. [234] P.J.A. Frinking, N. de Jong, Acoustic modeling of shell-encapsulated gas bubbles, Ultrasound Med Biol, 24 (1998) 523-533. [235] A. Bouakaz, P.J. Frinking, N. de Jong, N. Bom, Noninvasive measurement of the hydrostatic pressure in a fluid-filled cavity based on the disappearance time of micrometer-sized free gas bubbles, Ultrasound Med Biol, 25 (1999) 1407-1415. [236] V. Sboros, C.A. MacDonald, S.D. Pye, C.M. Moran, J. Gomatam, W.N. McDicken, The dependence of ultrasound contrast agents backscatter on acoustic pressure: theory versus experiment, Ultrasonics, 40 (2002) 579-583. [237] C.C. Church, The Effects of an Elastic Solid-Surface Layer on the Radial Pulsations of Gas- Bubbles, Journal of the Acoustical Society of America, 97 (1995) 1510-1521. [238] H.J. Vos, B. Dollet, J.G. Bosch, M. Versluis, N. de Jong, Nonspherical Vibrations of Microbubbles in Contact with a Wall—A Pilot Study at Low Mechanical Index, Ultrasound Med Biol, 34 (2008) 685- 688. [239] L.A. Crum, Measurements of the growth of air bubbles by rectified diffusion, The Journal of the Acoustical Society of America, 68 (1980) 203-211. [240] W. Nichols, M. O'Rourke, C. Vlachopoulos, McDonald's Blood Flow in Arteries, 6 ed., CRC Press2001. [241] W.C. Moss, Understanding the periodic driving pressure in the Rayleigh--Plesset equation, The Journal of the Acoustical Society of America, 101 (1997) 1187-1190. [242] J.S. Allen, R.A. Roy, Dynamics of gas bubbles in viscoelastic fluids. II. Nonlinear viscoelasticity, Journal of the Acoustical Society of America, 108 (2000) 1640-1650. [243] Y.C. Fung, Biomechanics: Circulation, New York, 1997. [244] J.L. Bull, Cardiovascular bubble dynamics, Critical reviews in biomedical engineering, 33 (2005) 299-346. [245] C. Pozrikidis, Numerical simulation of the flow-induced deformation of red blood cells, Ann Biomed Eng, 31 (2003) 1194-1205. [246] D. Halpern, T.W. Secomb, The Squeezing of Red-Blood-Cells through Parallel-Sided Channels with near-Minimal Widths, J Fluid Mech, 244 (1992) 307-322. [247] T.W. Secomb, R. Hsu, A.R. Pries, Blood flow and red blood cell deformation in nonuniform capillaries: Effects of the endothelial surface layer, Microcirculation, 9 (2002) 189-196. [248] X.J. He, D.N. Ku, Pulsatile flow in the human left coronary artery bifurcation: Average conditions, J Biomech Eng-T Asme, 118 (1996) 74-82. [249] D.N. Ku, Blood flow in arteries, Annu Rev Fluid Mech, 29 (1997) 399-434. [250] D.A. Edwards, H. Brenner, D.T. Wasan, Interfacial Transport Processes and Rheology, Butterworth-Heinemann, Boston, MA, 1991. [251] S.P. Evgenidis, N.A. Kazakis, T.D. Karapantsios, Bubbly flow characteristics during decompression sickness: Effect of surfactant and electrolyte on bubble size distribution, Colloid Surface A, 365 (2010) 46-51. [252] D.M. Eckmann, S.L. Diamond, Surfactants attenuate gas embolism-induced thrombin production, Anesthesiology, 100 (2004) 77-84. [253] J. Hanwright, J. Zhou, G.M. Evans, K.P. Galvin, Influence of surfactant on gas bubble stability, Langmuir, 21 (2005) 4912-4920. 219

[254] in: K. Ackles (Ed.) Symposium on Blood-Bubble Interaction in Decompression Sickness, Defence R&D Canada (DRDC), previously DCIEM, Canada, December 1973. [255] C.A. Ward, D. McCullough, W.D. Fraser, Relation between complement activation and susceptibility to decompression sickness, J Appl Physiol, 62 (1987) 1160-1166. [256] A. Baz, S.I. Abdel-Khalik, Effect of intravascular bubbles on perfusate flow and gas elimination rates following simulated decompression of a model tissue, Undersea biomedical research, 13 (1986) 27-44. [257] J.R. Lindner, M.P. Coggins, S. Kaul, A.L. Klibanov, G.H. Brandenburger, K. Ley, Microbubble persistence in the microcirculation during ischemia/reperfusion and inflammation is caused by integrin- and complement-mediated adherence to activated leukocytes, Circulation, 101 (2000) 668- 675. [258] A. Suzuki, S.C. Armstead, D.M. Eckmann, Surfactant reduction in embolism bubble adhesion and endothelial damage, Anesthesiology, 101 (2004) 97-103. [259] J.L. Zakin, H.W. Bewersdorff, Surfactant drag reduction, Rev Chem Eng, 14 (1998) 253-320. [260] A.P. Sawchuk, J.L. Unthank, M.C. Dalsing, Drag reducing polymers may decrease atherosclerosis by increasing shear in areas normally exposed to low shear stress, J Vasc Surg, 30 (1999) 761-764. [261] S. Kobayashi, S.D. Crooks, D.M. Eckmann, In vitro surfactant mitigation of gas bubble contact- induced endothelial cell death, Undersea Hyperb Med, 38 (2011) 27-39. [262] R. Miller, Z. Policova, R. Sedev, A.W. Neumann, Relaxation Behavior of Human Albumin Adsorbed at the Solution Air Interface, Colloid Surface A, 76 (1993) 179-185. [263] E. Stride, The influence of surface adsorption on microbubble dynamics, Philos T R Soc A, 366 (2008) 2103-2115. [264] J.R. Lindner, S. Ismail, W.D. Spotnitz, D.M. Skyba, A.R. Jayaweera, S. Kaul, Albumin microbubble persistence during myocardial contrast echocardiography is associated with microvascular endothelial glycocalyx damage, Circulation, 98 (1998) 2187-2194. [265] D.M. Eckmann, D.P. Cavanagh, A.B. Branger, Wetting Characteristics of Aqueous Surfactant- Laden Drops, J Colloid Interf Sci, 242 (2001) 386-394. [266] T. Thorsen, H. Dalen, R. Bjerkvig, H. Holmsen, Transmission and scanning electron microscopy of N2 microbubble-activated human platelets in vitro, Undersea biomedical research, 14 (1987) 45- 58. [267] C. Marabotti, F. Chiesa, A. Scalzini, F. Antonelli, R. Lari, C. Franchini, P.G. Data, Cardiac and humoral changes induced by recreational scuba diving, Undersea & hyperbaric medicine : journal of the Undersea and Hyperbaric Medical Society, Inc, 26 (1999) 151-158. [268] A.A. Bove, J.M. Hallenbeck, D.H. Elliott, Changes in blood and plasma volumes in dogs during decompression sickness, Aerospace medicine, 45 (1974) 49-55. [269] K. Yanagisawa, F. Moriyasu, T. Miyahara, M. Yuki, H. Iijima, Phagocytosis of ultrasound contrast agent microbubbles by Kupffer cells, Ultrasound in medicine & biology, 33 (2007) 318-325. [270] S. Milo, E. Rambod, C. Gutfinger, M. Gharib, Mitral mechanical heart valves: in vitro studies of their closure, vortex and microbubble formation with possible medical implications, Eur J Cardio- Thorac, 24 (2003) 364-370. [271] P. Cordes, R. Keil, T. Bartsch, K. Tetzlaff, M. Reuter, A. Hutzelmann, L. Friege, T. Meyer, E. Bettinghausen, G. Deuschl, Neurologic outcome of controlled compressed-air diving, Neurology, 55 (2000) 1743-1745. [272] C.L. Taylor, J.I. Macdiarmid, J.A. Ross, L.M. Osman, S.J. Watt, W. Adie, J.R. Crawford, A. Lawson, Objective neuropsychological test performance of professional divers reporting a subjective complaint of "forgetfulness or loss of concentration", Scandinavian journal of work, environment & health, 32 (2006) 310-317. [273] R. Macarez, Y. Dordain, M. Hugon, J.L. Kovalski, B. Guigon, S. Bazin, F. May, J. Colin, [Long-term effects of iterative diving on visual field, color vision and contrast sensitivity in professional divers], Journal francais d'ophtalmologie, 28 (2005) 825-831. 220

[274] P. Wilmshurst, C.J. Edge, P. Bryson, Long-term adverse effects of scuba diving, Lancet, 346 (1995) 384. [275] D.H. Elliott, J.A. Harrison, Bone necrosis--an occupational hazard of diving, Journal of the Royal Naval Medical Service, 56 (1970) 140-161. [276] W. Hemelryck, P. Germonpré, V. Papadopoulou, M. Rozloznik, C. Balestra, Long term effects of recreational SCUBA diving on higher cognitive function, Scandinavian journal of medicine & science in sports, 24 (2014) 928-934. [277] D. McQueen, G. Kent, A. Murrison, Self-reported long-term effects of diving and decompression illness in recreational scuba divers, British journal of sports medicine, 28 (1994) 101- 104. [278] J.M. Pontier, F. Guerrero, O. Castagna, Bubble Formation and Endothelial Function Before and After 3 Months of Dive Training, Aviat Space Envir Md, 80 (2009) 15-19. [279] Q. Wang, F. Guerrero, A. Mazur, K. Lambrechts, P. Buzzacott, M. Belhomme, M. Theron, Reactive Oxygen Species, Mitochondria, and Endothelial Cell Death during In Vitro Simulated Dives, Medicine and science in sports and exercise, 47 (2015) 1362-1371. [280] K. Lambrechts, J.M. Pontier, C. Balestra, A. Mazur, Q. Wang, P. Buzzacott, M. Theron, J. Mansourati, F. Guerrero, Effect of a single, open-sea, air scuba dive on human micro- and macrovascular function, Eur J Appl Physiol, 113 (2013) 2637-2645. [281] A.A. Buhlmann, P. Frei, H. Keller, Saturation and desaturation with N2 and He at 4 atm, J Appl Physiol, 23 (1967) 458-462. [282] R. Hamilton, R. Rogers, M. Powell, R. Vann, R. Dunford, M. Spencer, D. Richardson, The DSAT Recreational Dive Planner: Development and Validation of No-Stop Decompression Procedures for Recreational Diving, SPUMS J, 24 (1994) 206. [283] R.G. Dunford, P.J. Denoble, R.D. Vann, J.S. Shannon, N.W. Pollock, L.E. Howle, The relationship of age and BMI to Doppler-detected bubble grades., Undersea and Hyperbaric Medical Society 2008 Annual Scientific MeetingSalt Lake City Marriott Downtown, Salt Lake City, Utah., 2008. [284] R. Walker, Women in diving, SPUMS Journal, 26 (1996) 34-39. [285] A.A. Pilmanis, Ascent and silent bubbles, in: M.A. Lang, G.H. Egstrom (Eds.) Proceedings of the biomechanics of safe ascents workshop, American Academy of Underwater Sciences, Woods Hole, Massachusetts, 1989. [286] P.B. Bennett, A. Marroni, C. Balestra, R. Cali Corleo, P. Germonpré, M. Pieri, C. Bonuccelli, What ascent profile for the prevention of decompression sickness? 1. Recent research on the Hill/Haldane ascent controversy., Proceedings of the 28th Annual Scientific Meeting of the European Underwater and Biomedical Society. Pp 35-38:2002. September 4-8, Brugge, Belgium. [287] D. Carturan, A. Boussuges, G. Habib, B. Gardette, J.M. Sainty, Influence of ascent rate on venous bubbles detected after recreational dives, in: A. Marroni, G. Oriani (Eds.) Proc. Internat. Joint Meeting on Hyperbaric and Underwater Med.Milan. [288] T.S. Neuman, D.A. Hall, P.G. Linaweaver, Jr., Gas phase separation during decompression in man: ultrasound monitoring, Undersea Biomed Res, 3 (1976) 121-130. [289] D.M. Uguccioni, R.D. Vann, L.R. Smith, B.D. Butler, D.B. Rove, R.D. Roer, Effect of safety stops on venous gas emboli after no-stop diving, Undersea Hyperbar M, 22 (Suppl) (1995). [290] D. Carturan, A. Boussuges, P. Vanuxem, A. Bar-Hen, H. Burnet, B. Gardette, Ascent rate, age, maximal oxygen uptake, adiposity, and circulating venous bubbles after diving, J Appl Physiol (1985), 93 (2002) 1349-1356. [291] G.W. Pollard, P.L. Marsh, C.E. Fife, L.R. Smith, R.D. Vann, Ascent rate, post-dive exercise, and decompression sickness in the rat, Undersea Hyperb Med, 22 (1995) 367-376. [292] D. Vote, R. Vann, DAN USA Ascent Rate Study, South Pacific Underwater Medicine Society (SPUMS) Journal, 33 (2003). [293] A. Davis, Best Ascent Speed for Scuba Diving (accessed 07/09/2014). Available at: http://scubatechphilippines.com/scuba_blog/best-ascent-speed-scuba-diving/.

221

[294] O. Hyldegaard, M. Moller, J. Madsen, Protective Effect of Oxygen and Heliox Breathing during Development of Spinal Decompression-Sickness, Undersea Hyperbar M, 21 (1994) 115-128. [295] O. Hyldegaard, J. Madsen, Effect of Air, Heliox, and Oxygen Breathing on Air Bubbles in Aqueous Tissues in the Rat, Undersea Hyperbar M, 21 (1994) 413-424. [296] O. Hyldegaard, J. Madsen, Effect of hypobaric air, oxygen, heliox (50 : 50), or heliox (80 : 20) breathing on air bubbles in adipose tissue, J Appl Physiol, 103 (2007) 757-762. [297] O. Hyldegaard, M. Moller, J. Madsen, Effect of He-O2, O2, and N2o-O2 Breathing on Injected Bubbles in Spinal White Matter, Undersea Biomed Res, 18 (1991) 361-371. [298] R. Arieli, A. Marmur, Dynamics of gas micronuclei formed on a flat hydrophobic surface, the predecessors of decompression bubbles, Resp Physiol Neurobi, 185 (2013) 647-652. [299] D. Carturan, A. Boussuges, H. Burnet, J. Fondarai, P. Vanuxem, B. Gardette, Circulating venous bubbles in recreational diving: relationships with age, weight, maximal oxygen uptake and body fat percentage, Int J Sports Med, 20 (1999) 410-414. [300] C. Acott, D. Doolette, Simulations of near-drowning and decompression sickness: A preliminary study, (2002). [301] J.-E. Blatteau, J. Hugon, E. Gempp, O. Castagna, C. Pény, N. Vallée, Oxygen breathing or recompression during decompression from nitrox dives with a rebreather: effects on intravascular bubble burden and ramifications for decompression profiles, Eur J Appl Physiol, 112 (2012) 2257- 2265. [302] E.R. Davies, A modified Hough scheme for general circle location, Pattern Recognition Letters, 7 (1988) 37-43. [303] C. Kimme, D. Ballard, J. Sklansky, Finding circles by an array of accumulators, Commun. ACM, 18 (1975) 120-122. [304] H. Li, M.A. Lavin, R.J. Le Master, Fast Hough transform: A hierarchical approach, Computer Vision, Graphics, and Image Processing, 36 (1986) 139-161. [305] L.G. Leal, Advanced Transport Phenomena: Fluid Mechanics and Convective Transport Properties, Cambridge University Press, Cambridge, 2007. [306] W.M. Buehl, J.W. Westwater, Bubble growth by dissolution: Influence of contact angle, Aiche J, 12 (1966) 571-576. [307] S.F. Jones, K.P. Galvin, G.M. Evans, G.J. Jameson, Carbonated water: The physics of the cycle of bubble production, Chem Eng Sci, 53 (1998) 169-173. [308] G.S. Barker, B. Jefferson, S.J. Judd, The control of bubble size in carbonated beverages, Chem Eng Sci, 57 (2002) 565-573. [309] R. Arieli, Was the appearance of surfactants in air breathing vertebrates ultimately the cause of decompression sickness and autoimmune disease?, Respir Physiol Neurobiol, 206c (2015) 15-18. [310] C.G.J. Bisperink, A. Prins, Bubble-Growth in Carbonated Liquids, Colloid Surface A, 85 (1994) 237-253. [311] M.P. Spencer, A. Johanson, Investigation of New Principles for Human Decompression Schedules Using Doppler Ultrasound Blood Bubble Detection. Technical Report to ONR on contract N00014-73-C-0094, Seattle: Institute for Environmental Medicine and Physiology, 1974. [312] K. Kisman, G. Masurel, Method for evaluating circulating bubbles detected by means of the Doppler ultrasonic method using the 'K.M. code' (English translation of 283 CERTSM 1983), Toulon: Centre d'Etudes et de Recherches Techniques Sous-Marines, 1983. [313] K.D. Sawatzky, R.Y. Nishi, Assessment of inter-rater agreement on the grading of intravascular bubble signals, Undersea Biomed Res, 18 (1991) 373-396. [314] K.D. Sawatzky, The relationship between intravascular Doppler-detected gas bubbles and decompression sickness after bounce diving in humans, Toronto, ON: York University, 1991. [315] A. Marroni, P.B. Bennett, F.J. Cronje, R. Cali-Corleo, P. Germonpre, M. Pieri, C. Bonuccelli, C. Balestra, A deep stop during decompression from 82 fsw (25 m) significantly reduces bubbles and fast tissue gas tensions, Undersea Hyperb Med, 31 (2004) 233-243.

222

[316] P.B. Bennett, A. Marroni, F.J. Cronje, R. Cali-Corleo, P. Germonpre, M. Pieri, C. Bonuccelli, M.G. Leonardi, C. Balestra, Effect of varying deep stop times and shallow stop times on precordial bubbles after dives to 25 msw (82 fsw), Undersea Hyperb Med, 34 (2007) 399-406. [317] O.S. Eftedal, S. Lydersen, A.O. Brubakk, The relationship between venous gas bubbles and adverse effects of decompression after air dives., Undersea Hyperb Med, 34 (2007) 99-105. [318] O. Eftedal, A.O. Brubakk, Agreement between trained and untrained observers in grading intravascular bubble signals in ultrasonic images, Undersea Hyperb Med, 24 (1997) 293-299. [319] S. Choudhry, B. Gorman, J.W. Charboneau, D.J. Tradup, R.J. Beck, J.M. Kofler, D.S. Groth, Comparison of tissue harmonic imaging with conventional US in abdominal disease, Radiographics, 20 (2000) 1127-1135. [320] T. Uppal, Tissue Harmonic Imaging, AJUM, 13 (2010) 29-31. [321] C. Daniels, C. Weytjens, B. Cosyns, D. Schoors, J. De Sutter, B. Paelinck, L. Muyldermans, G. Van Camp, Second harmonic transthoracic echocardiography: the new reference screening method for the detection of patent foramen ovale, Eur J Echocardiogr, 5 (2004) 449-452. [322] O. Eftedal, PhD Thesis: Ultrasonic detection of decompression induced vascular microbubbles, NTNU, Trondheim, Norway, 2007. [323] N. Dejong, L. Hoff, T. Skotland, N. Bom, Absorption and Scatter of Encapsulated Gas Filled Microspheres - Theoretical Considerations and Some Measurements, Ultrasonics, 30 (1992) 95-103. [324] B.C. Eatock, R.Y. Nishi, G.W. Johnston, Numerical-Studies of the Spectrum of Low-Intensity Ultrasound Scattered by Bubbles, Journal of the Acoustical Society of America, 77 (1985) 1692-1701. [325] J. Cohen, Weighted kappa: nominal scale agreement with provision for scaled disagreement or partial credit, Psychol Bull, 70 (1968) 213-220. [326] J. Cohen, A coefficient of agreement for nominal scales, Educ Psychol Meas, 20 (1960) 207-216. [327] H.C. Kraemer, Extension of the Kappa-Coefficient, Biometrics, 36 (1980) 207-216. [328] J.M. Bland, D.G. Altman, Statistical Methods for Assessing Agreement between Two Methods of Clinical Measurement, Lancet, 1 (1986) 307-310. [329] J.R. Landis, G.G. Koch, An application of hierarchical kappa-type statistics in the assessment of majority agreement among multiple observers, Biometrics, 33 (1977) 363-374. [330] J.R. Landis, G.G. Koch, The measurement of observer agreement for categorical data, Biometrics, 33 (1977) 159-174. [331] V. Rousson, T. Gasser, B. Seifert, Assessing intrarater, interrater and test-retest reliability of continuous measurements, Stat Med, 21 (2002) 3431-3446. [332] P.E. Shrout, J.L. Fleiss, Intraclass correlations: uses in assessing rater reliability, Psychol Bull, 86 (1979) 420-428. [333] J.L. Fleiss, J. Cohen, Equivalence of Weighted Kappa and Intraclass Correlation Coefficient as Measures of Reliability, Educ Psychol Meas, 33 (1973) 613-619. [334] J.L. Fleiss, P.E. Shrout, Approximate Interval Estimation for a Certain Intraclass Correlation- Coefficient, Psychometrika, 43 (1978) 259-262. [335] J.M. Bland, D.G. Altman, Statistical methods for assessing agreement between two methods of clinical measurement, Int J Nurs Stud, 47 (2010) 931-936. [336] K.E. Kisman, G. Masurel, R. Guillerm, Bubble Evaluation Code for Doppler Ultrasonic Decompression Data, Undersea Biomed Res, 5 (1978) 28. [337] R.Y. Nishi, K.E. Kisman, B.C. Eatock, I.P. Buckingham, M. G., Assessment of decompression profiles and divers by Doppler ultrasonic monitoring, in: A.J. Bachrach, M.M. Matzen (Eds.) Underwater Physiology VII: Proceedings of the Seventh Symposium on Underwater PhysiologyBethesda MD: Undersea Medical Society, 1981, pp. 717-727. [338] K. Kisman, G. Masurel, D. LaGrue, J. Le Pêchon, [Evaluation of the quality of decompression using ultrasound bubble detection]. French, Méd Aéro Spat Méd Sub Hyp, 67 (1978) 293-297. [339] V. Papadopoulou, J. Hui, C. Balestra, W. Hemelryck, P. Germonpre, R. Eckersley, M. Tang, Evaluating Counting of Venous Gas Emboli on Post-SCUBA dive echocardiographs, ID 464, 2013 IEEE Joint UFFC, EFTF and PFM SymposiumPrague, 2013. 223

[340] A.G. Fredriksson, J. Zajac, J. Eriksson, P. Dyverfeldt, A.F. Bolger, T. Ebbers, C.J. Carlhall, 4-D blood flow in the human right ventricle, American journal of physiology. Heart and circulatory physiology, 301 (2011) H2344-2350. [341] L. Wigstrom, T. Ebbers, A. Fyrenius, M. Karlsson, J. Engvall, B. Wranne, A.F. Bolger, Particle trace visualization of intracardiac flow using time-resolved 3D phase contrast MRI, Magnet Reson Med, 41 (1999) 793-799. [342] D. Bakovic, D. Glavas, I. Palada, T. Breskovic, D. Fabijanic, A. Obad, Z. Valic, A.O. Brubakk, Z. Dujic, High-grade bubbles in left and right heart in an asymptomatic diver at rest after surfacing, Aviat Space Environ Med, 79 (2008) 626-628. [343] F.J. Romero-Bermejo, M. Ruiz-Bailen, M. Guerrero-De-Mier, J. Lopez-Alvaro, Echocardiographic hemodynamic monitoring in the critically ill patient, Curr Cardiol Rev, 7 (2011) 146-156. [344] B.D. Butler, C. Fife, T. Sutton, M. Pogodsky, P. Chen, Hepatic portal venous gas with hyperbaric decompression: ultrasonographic identification, Journal of ultrasound in medicine : official journal of the American Institute of Ultrasound in Medicine, 14 (1995) 967-970. [345] N. Bird, CT finding of VGE in the portal veins and IVC in a diver with abdominal pain: a case report, Undersea Hyperb Med, 34 (2007) 393-397. [346] V. Papadopoulou, R.J. Eckersley, C. Balestra, T.D. Karapantsios, M.X. Tang, A critical review of physiological bubble formation in hyperbaric decompression, Adv Colloid Interfac, 191 (2013) 22-30. [347] E. Gempp, J.E. Blatteau, J.M. Pontier, C. Balestra, P. Louge, Preventive effect of pre-dive hydration on bubble formation in divers, British journal of sports medicine, 43 (2009) 224-228. [348] P. Germonpre, F. Hastir, P. Dendale, A. Marroni, A.F. Nguyen, C. Balestra, Evidence for increasing patency of the foramen ovale in divers, The American journal of cardiology, 95 (2005) 912-915. [349] M. Rozloznik, W. Hemelryck, F. Tillmans, V. Papadopoulou, S. Theunissen, C. Balestra, Effect of Pre-Dive Hydration on Venous Gas Bubble Production in Divers, Presented at the Hyperbaric and Aerospace Medicine Congress, Prague, Czech Republic, 23-24 May 2013. [350] S.P. Evgenidis, T.D. Karapantsios, Effect of bubble size on void fraction fluctuations in dispersed bubble flows, Int J Multiphas Flow, 75 (2015) 163-173. [351] R.G. Dunford, R.D. Vann, W.A. Gerth, C.F. Pieper, K. Huggins, C. Wacholtz, P.B. Bennett, The incidence of venous gas emboli in recreational diving, Undersea Hyperb Med, 29 (2002) 247-259. [352] M. Gennser, K.M. Jurd, S.L. Blogg, Pre-dive exercise and post-dive evolution of venous gas emboli, Aviat Space Environ Med, 83 (2012) 30-34. [353] J.G. Swan, J.C. Wilbur, K.L. Moodie, S.A. Kane, D.A. Knaus, S.D. Phillips, T.L. Beach, A.M. Fellows, P.J. Magari, J.C. Buckey, Microbubbles are detected prior to larger bubbles following decompression, J Appl Physiol (1985), 116 (2014) 790-796. [354] H. Fok, B. Jiang, P. Chowienczyk, B. Clapp, Microbubbles shunting via a patent foramen ovale impair endothelial function, JRSM Cardiovascular Disease, 4 (2015) 2048004015601564. [355] G. Carneiro, J.C. Nascimento, A. Freitas, The Segmentation of the Left Ventricle of the Heart From Ultrasound Data Using Deep Learning Architectures and Derivative-Based Search Methods, Ieee T Image Process, 21 (2012) 968-982. [356] J.W. Klingler, C.L. Vaughan, T.D. Fraker, L.T. Andrews, Segmentation of Echocardiographic Images Using Mathematical Morphology, Ieee T Bio-Med Eng, 35 (1988) 925-934. [357] D.B. Sher, S. Revankar, S. Rosenthal, Computer methods in quantitation of cardiac wall parameters from two dimensional echocardiograms: a survey, International journal of cardiac imaging, 8 (1992) 11-26. [358] A.A. Barath, Image Processing Lecture Notes, Imperial College Bioengineering (BE3-HIPR), 2013. [359] N.M. Dhutia, G.D. Cole, K. Willson, D. Rueckert, K.H. Parker, A.D. Hughes, D.P. Francis, A new automated system to identify a consistent sampling position to make tissue Doppler and transmitral Doppler measurements of E, E ' and E/E ', Int J Cardiol, 155 (2012) 394-399.

224

[360] M. Zolgharni, N.M. Dhutia, G.D. Cole, K. Willson, D.P. Francis, Feasibility of using a reliable automated Doppler flow velocity measurements for research and clinical practices, Proc Spie, 9040 (2014).

225