GRADUATE SCHOOL OF BIOMEDICAL ENGINEERING
Simulation of Aortic Valve Dynamics during Left Ventricular Support
Khalid.A.Alonazi
B.BiomedE, King Saud University M.BiomedE, University of New South Wales
A dissertation submitted for the degree of Doctor of Philosophy
February 2015
ORIGINALITY STATEMENT
‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at
UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’
Signed ……………………………………………......
Date ……………………………………………......
AUTHENTICITY STATEMENT
‘I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.’
Signed ……………………………………………......
Date ……………………………………………......
Table Contents
Acknowledgments ...... i Abstract ...... iii List of Abbreviations ...... iv List of Figures ...... vi List of Tables ...... xi Part I-Introduction, Background and Literature Review ...... 1 1 Chapter 1 - Introduction, Aims, and Outline ...... 2 1.1 Research Motivation ...... 2 1.2 Thesis Aims ...... 4 1.3 Thesis outline ...... 6 1.4 Publications ...... 9 1.4.1 Refereed Conference Proceedings and Abstracts ...... 9 2 Chapter 2 - Background...... 11 2.1 Overview ...... 11 2.2 Basic Anatomy and Physiology ...... 11 2.2.1 Cardiovascular system ...... 11 2.2.2 Control of Cardiac Output ...... 16 2.2.3 Aortic Valve Anatomy and Physiology ...... 19 2.2.4 Regulation of the Aortic Valve ...... 21 2.3 Heart Failure Pathology and Therapy ...... 24 2.3.1 Congestive Heart Failure ...... 24 2.3.2 Ventricular Assist Devices: an Emerging Therapy for CHF ...... 28 2.3.3 Abnormal pathophysiology in LVAD patients ...... 40 2.3.4 LVAD control systems ...... 40 2.4 Existing LVAD Models and Computational Methods ...... 54 2.4.1 Overview of Computational Techniques for LVAD-AV Modelling ...... 54 2.4.2 Finite Element Method ...... 56 2.4.3 Heart-Pump Interaction Models ...... 58
3 Chapter 3-Review of Modelling Approaches in Cardiac Dynamics ...... 60 3.1 Background ...... 60 3.2 Existing cardiac dynamics models ...... 63 3.3 Existing modelling studies of AV state under LVAD support ...... 68 Part II – Methods ...... 71 4 Chapter 4 -Development of Computational 2D LV-pump Models ...... 72 4.1 Introduction ...... 72 4.2 Geometrical models of LV-LVAD interaction ...... 75 4.2.1 Simplified LV-pump interaction model ...... 75 4.2.2 Simplified LV-pump interaction model with systemic circulation...... 83 4.2.3 Simplified LV-pump model with systemic circulation and heart contractility .. 89 4.2.4 Simplified LV-pump interaction model with systemic circulation, cardiac contractility and without AV ...... 95 4.2.5 Realistic LV-Pump geometry model with and without AV leaflets ...... 103 4.2.6 Simulating LV and AV dynamics ...... 119 4.2.7 Investigation of Cardiovascular Interaction with a Left ventricular Assist Device 120 Part III-Results and Discussion ...... 124 5 Chapter 5 - Fluid-Structure Interaction in a Simple Model of an Assisted Left Ventricle ...... 125 5.1 Introduction ...... 125 5.2 Simplified LV-pump model ...... 126 5.2.1 Results: ...... 126 5.2.2 Discussion ...... 128 5.3 Simplified LV-pump model with systemic circulation ...... 129 5.3.1 Results: ...... 129 5.3.2 Discussion ...... 139 5.4 Conclusion ...... 141 6 Chapter 6 - Simulation of Aortic Valve Response during Ventricular Assist Device Support ...... 142 6.1 Introduction ...... 142 6.2 Simplified LV-pump model ...... 143 6.3 Realistic LV-Pump model with AV leaflets ...... 147
6.3.1 Identification of aortic valve states for LV- pump interaction models with AV 148 6.3.2 Detection of AV State ...... 150 6.3.3 Discussion ...... 161 6.4 Limitations ...... 162 6.5 Conclusion ...... 163 7 Chapter 7 – Effect of Parameter Variations on Aortic Valve State under Rotary Blood Pump Assistance...... 165 7.1 Introduction ...... 165 7.2 Simplified LV-pump model ...... 168 7.2.1 Results and Discussion ...... 168 7.3 Realistic LV-Pump model ...... 174 7.3.1 Introduction ...... 174 7.3.2 Results: ...... 174 7.3.3 Discussion ...... 181 7.4 Model Limitations ...... 183 7.5 Conclusions ...... 183 8 Chapter 8 - Simulation of Motor Current Waveforms as an Index for Aortic Valve Condition during Ventricular Support ...... 184 8.1 Introduction ...... 185 8.2 Simplified LV-pump model ...... 187 8.2.1 Results: ...... 187 8.2.2 Discussion ...... 190 8.3 Realistic LV-Pump model Results ...... 192 8.3.1 Identification of AV state ...... 192 8.3.2 LV wall motion ...... 193 8.3.3 LV-Pump model simulations ...... 194 8.3.4 Discussion ...... 204 8.4 Model Limitations ...... 205 8.5 Conclusions ...... 206 9 Chapter 9 - Conclusions and Recommendations ...... 207 9.1 Conclusions ...... 207 9.2 Suggestions for Future Work ...... 209 9.2.1 Simulating chronic heart failure during exercise and postural change ...... 210
9.2.2 AV state detection validation ...... 210 9.2.3 Effect of blood viscosity on AV states ...... 211 Bibliography ...... 212
Acknowledgments
Writing these acknowledgements I look back and remember the time through the years I have spent in my Phd. Although the amount of studies never ends, it was a great experience. Praise
God for the blessing of science. Thanks to my professors and friends. For me it was the last stage of my long schooling life, but I will not stop in my research journey to serve my society and all of humanity. I would like to thank the Custodian of the two Holy Mosques King
Abdullah bin Abdul Aziz, the King of Saudi Arabia, for his support to Saudi students all over the world, as well as the Ministry of Defence representative, my commander Major General
Walid Khalil. I would also like to thank my mother for her patience, love, support, and encouragement. I would like to thank all those inspiring persons I have learnt from. In particular, I would like to thank all of my teachers who have taught me over the last 4 years, and all my friends in Australia and abroad for sharing this journey with me. In specific, my supervisors Professor Socrates Dokos, Professor Nigel Lovell, and Professor Andrey Savkin, because of their knowledge, guidance, assistance and follow-up to the completion of my thesis, giving me the opportunity to present my research at international conferences, and for giving me the confidence in myself to complete my research in the simulation of heart dynamics. Thank you Professor Socrates Dokos for your help in all aspects of this project, encouraging words at difficult times for the completion of my thesis. Thanks, Professor Nigel
Lovell for your advice. I wish to thank my friend Associate Professor Abdul Hakim for giving me a concrete base in the heart pump control systems. Special thanks to my friends
Amr, Mohammad, Fahad, Ammar, James, Adrian, Einly, Tianruo, Siwei ,Azam and Ulises.
i
thank you guys for your support. Finally, my gratitude is also extended to my family in Saudi
Arabia, my sisters and my brothers: thank you for all your support, encouragement and patience - without you this PhD would not have happened.
ii
Abstract
Implantable Rotary Blood Pumps (IRBPs) for the left ventricle (LV) have become a
viable treatment and long-term option for heart failure (HF) patients. In addition,
development of valve abnormalities after Left Ventricular Assist Device (LVAD)
implantation is common among patients with advanced HF, likely due the fact that
the LVAD alters haemodynamics by changing the direction of blood flow from the
apex of the heart, largely bypassing the left ventricle (LV), directly to the aorta. The
aim of this thesis was to investigate the hemodynamic interaction between the
LVAD and aortic valve (AV) using 2D LV-LVAD computational models.
To investigate the correlation between AV status and LVAD motor speed, this study
perfomed a detailed computational analysis of left ventricular flow and mechanics
during LVAD support. Simulations were carried out using a 2D LV-pump Fluid-
Structure Interaction (FSI) approach, examining LVAD intrinsic motor current and
motor speed waveforms. AV state was assessed by analysing the pump motor
current waveform, investigating its association with open-close valve state and pump
impeller speed. Results show that there is a significantly higher motor current during
the valve open state, which has the potential of being utilized in future LVAD
control systems to ensure patient safety and comfort, and reduce the incidence of AV
pathologies during heart pump support, as well as for optimal management of pump
outflow, paving the way for more sophisticated pump control algorithms, which take
into account heart valve state.
iii
List of Abbreviations
AC aortic valve open ECG electrocardiogram
ACE angiotensin converting enzyme EDV end-diastolic volume
AI aortic insufficiency ESV end-systolic volume
AR aortic regurgitation EDP end-diastolic pressure
AS aortic stenosis ESP end- systolic pressure
AO aortic valve closed FDA food and drug administration
AG aqueous glycerol FD fictitious domain
ALE Arbitrary Lagrangian Eulerian FE finite element
ALI activity level index FSI fluid structure interaction
ANO aortic valve not open H pump head
AP arterial pressure HCT haematocrit
AoP aortic pressure HR heart rate
ARX autoregressive with exogenous IRBP implantable rotary blood pump input LV left ventricle
AV aortic valve LVAD left ventricular assist device
CHF congestive heart failure LVP left ventricular pressure
CO cardiac output MV mitral valve
CFD computational fluid dynamics NYHA New York Heart Association method PV pulmonary valve
CHF congestive heart failure rpm pump rotational speed
CVS cardiovascular system RV right ventricle
iv
SOV2 mixed venous oxygen saturation Q flow rate
WHO world health organization Qa aortic flow rate
ΔP pump differential pressure Qav aortic valve flow rate
SV stroke volume Qp pump flow rate
SVR systemic vascular resistance rpm revolutions per minute
TAH total artificial heart ω pump impeller speed
VAD ventricular assist device I pump electrical current
VI electrical input power 2D two dimensional
Vtotal inlet-flow rate to LV model.
v
List of Figures
Figure 2.1:A longitudinal view of the heart and its main components (adapted ...... 12 Figure 2.2: Distribution of blood volume in the different parts of the ...... 13 Figure 2.3: “Syncytial” interconnecting nature of cardiac muscle fibers (adapted from Guyton et al. [12])...... 14 Figure 2.4 : Ventricular pumping action; systole and diastole (adapted from Mohrman et al. [13])...... 14 Figure 2.5: Left ventricular function during cardiac cycle illustrating variations in left atrial pressure, left ventricular pressure, aortic pressure, pressure gradient, the electrocardiogram (ECG), and the phonocardiogram (adapted from Yellin et al. [14]). 15 Figure 2.6 : Heart pump function. The gray arrow indicates the influence of increased ventricular filling, while the black arrows indicate the findings of Frank and Starling .. 17 Figure 2.7 : Typical pressure-volume relationship of the left ventricle. The letters indicate valve action: A. mitral valve (MV) opens; B. mitral valve closes; C. aortic valve opens; D. aortic valve closes. (adapted from Vandenberghe [15])...... 18 Figure 2.8 : Schematic representation of the aortic valve: (a) side view of the complete valve, (b) after dissection of one leaflet with corresponding sinus wall and, (c) aortic view, (ada pted ...... 20 Figure 2.9 : Pressure and flow curves for aortic and mitral valves (adapted from Yoganathan et al. [21])...... 22 Figure 2.10: Aortic (Pao: dashed line) and left ventricular (Plv: solid line) pressure curves during the cardiac cycle. The associated flow curve is also given (Q: dotted line). AO denotes the onset of valve opening and AC the moment of complete closure (adapted from Yoganathan et al. [22])...... 23 Figure 2.11 : Cardiovascular alterations with compensated systolic heart failure. Point (A) illustrates the intersection of normal CO and normal venous function curves (adapted from Alomari et.al [13])...... 26 Figure 2.12 : Left ventricular pressure-volume loops during heart failure ...... 27 Figure 2.13 : Prof. Christiaan Barnard with the first recipient of a donor heart (adapted from Vandenberghe [15])...... 30 Figure 2.14: Schematic of the myocardial wedge excised from the left ventricle in the Batista procedure. The resulting exposed edges of the ventricle are subsequently sewn together to yield a ‘remodelled’ ventricle (adapted from Starling et al. [38])...... 32 Figure 2.15 : Pulsatile-Flow (Panel A) and continuous-Flow (Panel B) Left Ventricular Assist Devices (LVADs) (adapted from Slaughter et al. [5])...... 37 Figure 2.16 : VentrAssistTM pump ( adapted from Gosline [17])...... 38 Figure 4.1 : Simplified 2D representation of the LV, aortic valve ...... 76
vi
Figure 4.2 : (a) a fluid and structural element meshes in the simplified LV-pump model. The outer and inner AV segment lengths are 13 and 15.3 mm, respectively. (b) zoomed view of the AV highlighting the leaflet thickness, AV mesh and chamber element details. ... 78 Figure 4.3 : Sinusoidal velocity profile multiplied by an instantaneous step starting at ...... 82 Figure 4.4 : (a) Fluid and structural element meshes of simplified LV with pump and CVS model. The length and diameter of the pump cannula are 0.6 and 0.6 mm, respectively. (b) Zoomed view of the AV highlighting the leaflet thickness, AV and chamber mesh element details...... 84 Figure 4.5 : Windkessel model of the circulation where is the left ventricular outlet pressure, is the arterial systemic pressure, is atrial inflow to the LV, is the pump flow rate (L/min), is the blood flow ejected from the LV, is the characteristic aortic impedance, is the peripheral resistance and is the arterial systemic compliance...... 85 Figure 4.6 : Sinusoidal velocity profile applied to the inlet boundaries to simulate diastolic inflow. The inflow velocity oscillated in a sinusoidal pattern with period 1 s and a magnitude 0.75ms−1 about a mean level of 0.25 ms-1...... 88 Figure 4.7 : (a) Fluid and structural domain meshes for the simplified LV-LVAD interaction model with heart wall contraction. The length and diameter of the pump cannula were 0.6 and 0.6 mm respectively. (b) Zoomed view of upper AV leaflet highlighting the leaflet domain and LV chamber mesh element details...... 90 Figure 4.8 : Windkessel model of the circulation where is the left ventricular outlet pressure, is the arterial systemic pressure, is atrial inflow to the LV, is the pump flow rate (L/min), is the blood flow ejected from the LV, is the characteristic aortic impedance, is the peripheral resistance, is the arterial systemic compliance and P is the pressure differential head of the LVAD pump...... 94 Figure 4.9 : 2D representation of the LV chamber and pump cannula without AV ...... 95 Figure 4.10 : (a) Fluid and structural domain meshes for the simplified LV model without AV. (b) Mesh element detail at the sink outflow boundary. (c) Zoomed view of theoutflow bo und-ary highlighting the mesh element detail around the aortic outflow...... 97 Figure 4.11 : Windkessel model used with simplified LV-pump model without AV leaflets...... 100 Figure 4.12 : 2D geometry of the LV and AV (long axis plane) with LVAD cannula attached at the apex. The diameter of the LVAD cannula was 0.87 cm...... 104 Figure 4.13 .Two-dimensional geometry of the LV (long axis plane), LA, and ...... 106 Figure 4.14 : (a) 2D realistic LV model mesh. (b) Zoomed-in view of boundary mesh layers at the interface between the LV walls and fluid. (c) Zoomed-in view of AV leaflet tips highlighting the mesh element detail...... 108 Figure 4.15 : Mesh element size in 2D realistic LV model showing the aorta (Ao) and AV leaflet regions. (a) Boundary layer sizes at the Ao (b) zoomed view of the AV leaflet tips highlighting the element sizes in mm...... 110 Figure 4.16 : (a) Element quality of realistic LV model (q) in the range from 0 to 1. (b) Zoomed-in view of boundary element quality at the interface between the LV wall and
vii
fluid. (c) Zoomed-in view of AV leaflet tips, highlighting the element quality around the AV...... 110 Figure 4.17 : 2D realistic representation of the LV, AV, pump cannula and ...... 113 Figure 4.18 : Two-dimensional realistic representation of the LV, pump cannula ...... 114 Figure 4.19 : Boundary conditions of realistic LV-pump model...... 118 Figure 5.1 : Simulated pressure at pump outlet over a single 1 s cardiac cycle, where ...... 127 Figure 5.2 : Snapshots of simulated LV blood velocity magnitude during LVAD support at128 Figure 5.3 : (a) Simulated pressure from the pump outlet over a single, 1 s cardiac cycle, where ' * ' and ' ● ' indicate aortic valve opening and closing times, respectively. ‘c’ represents the period during which the aortic valve is closed, and ‘o’ is the period in which the aortic valve is open. (b) Zoomed-in view of pumpoutlet pressure. T-he pressure shown at the position of the arrows increases transiently on aortic valve closure...... 130 Figure 5.4 : Snapshots of computed LV blood velocity magnitude during LVAD support at various phases during the ...... 132 Figure 6.1 : Simulated motor electric current and pump impeller speed at two motor speed set points (100 and 150 rad/s), where ' ■ ' and ' ● ' indicate AV opening and closing times, respectively, (a) Simulations using a motor speed set point of set =100 rad/s. Periods 'O' and 'C' represent the phases during which the AV is open and valve closed respectively. (b). Simulations using a motor speed set point of set =150 rad/s. Periods 'O' and 'C' represent the phases during which the AV is open and closed, respectively. Max PCT and Min PCT are the maximum and minimum values of current threshold during AV closure, respectively. The motor current waveform at Max PCT begins to decrease more rapidly once the aortic valve is closed at the closing notch (CN)...... 144 Figure 6.2 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle...... 146 Figure 6.3 : Simulated left ventricular pressure (sold black line) and aortic pressure (dotted149 Figure 6.4 : AV state in 2D LV-LVAD realistic geometry simulations. The left pa- ...... 150 Figure 6.5 : Simulated electric current and pump impeller speeds at four motor speed set points (50, 100, 150 and 180 rad/s), where ' ■ ' and ' ● ' indicate AV opening and closing times, respectively. Periods 'O' and 'C' represent the phases during which the AV is open, referred to as ventricular ejection (VE), and valve closed (VC), respectively. Max PCT and Min PCT are the maximum and minimum values of motor current during AV closure, respectively. The motor current waveform at Max PCT further decreases once the aortic valve is closed at the closing notch (CN). (a) set = 50 rad/s (b) set = 100 rad/s) (c)set = 150 rad/s) (d)set = 180 rad/s...... 154 Figure 6.6 : Simulated electric current (black line) and distance between aortic valve (AV) lea fl-ets (red line) at a motor speed set points of 200 rad/s, where '●' and '■' indicate AV o pening an-d closing times, respectively. The duration of the closing phase was increased from lower set p-oint values, with the VC state beingcontinuously maintained (ANO). A R denotes the aortic val-ve regurgitant period through the cardiac cycle...... 155
viii
Figure 6.7 : Average inflow rate Qin (dashed line) applied at the inlet (source) boundaries, and the inlet pump flow rate Qp (solid line)...... 155 Figure 6.8 : LVP–peak (red dotted line) and LVAD motor current ...... 157 Figure 6.9 : Closure time against motor speed in LV model ...... 157 Figure 6.10 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. For each panel, the three snapshots illustrate AV closing phase (left), opening phase (middle) and fully-open (right). (a) set = 50 rad/s. (b) set = 100 rad/s...... 159 Figure 6.11 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. For each panel, the three snapshots illustrate AV closing phase (left), opening phase (middle), and fully-open phase (right). (c) set = 150 rad/s. (d) set = 180 rad/s...... 160 Figure 6.12 :Average pump inflow rate (Qp) at a motor speed set points of 100 rad/s, where ' ●' and '■' indicate AV opening and closing times, respectively...... 162 Figure 7.1 : Simulated electric current and pump impeller speed at two motor speed set points of 100 rad/s (top) and150 rad/s (bottom), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively, the periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the periods during which the AV is open and closed, respectively at conditions H1, H2 and H3 respectively. MAX C1PT, MAX C2PT and MAX C3PT, and MIN C1PT, MIN C2PT and MIN C3PT are the maximum and minimum values of motor current during AV closure...... 169 Figure 7.2 : Simulated distance between AV leaflets at two motor speed set points of 100 rad/s (top) and150 rad/s (bottom), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. The periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent phases during which the AV is open and closed, respectively, for conditions H1, H2 and H3 respectively...... 171 Figure 7.3 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle for = 100 rad/s...... 173 Figure 7.4 : Simulated motor electric current at three motor speed set points (50, 100, and150 rad/s), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. The periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the phases during which the AV is open and closed respectively, at conditions H1, H2 and H3 respectively...... 175 Figure 7.5 : (a), (b) and (c). Simulated motor current waveform during AV closure/open phases, where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. Current levels MAX C1PT, MAX C2PT and MAX C3PT, and MIN C1PT, MIN C2PT and MIN C3PT denote the maximum and minimum values of motor current during AV closure, whilst MAX O1PT, MAX O2PT and MAX O3PT are the maximum current values during AV open phases. From top to bottom, the panels show pump speed set points of (a) set = 50 rad/s, (b) set 100 rad/s and (c) set = 150 rad/s)...... 177 Figure 7.6 : Simulated open and closed states of AV leaflets at two motor speed set points of (a) 50 rad/s and (b) 100 rad/s). The waveforms show the distance between AV leaflets, where the distance in the open state the distance was measured with a positive value in ix
millimeters. This open state is referred to as ventricular ejection (VE), whilst the valve closed (VC) state is when the distance was zero. Periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the phases during which the AV is open and closed, respectively at conditions H1, H2 and H3 respectively...... 178 Figure 7.7 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle for = 150 rad/s. (a) Aortic valve closing and opening phases under condition H1. (b) Aortic valve closing and opening phases under condition H2. (c) Aortic valve cl osing and opening phases under condition H3...... 180
Figure 8.1 : Simulated aortic valve flow (Qav), pump flow (Qp), left ventricular pressure (Plv) and aortic pressure (Pao), ...... 189 Figure 8.2 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. (a) = 50 rad/s. (b) = 100 rad/s. (c) = 150 rad/s...... 191 Figure 8.3 : Left ventricular wall simulated cavity dimension during contractile motion. ... 193 Figure 8.4 : Top panels: left ventricular pressure (black line) and aortic valve pressure (do tted red line) at variouspump speeds. Lower panels: ...... 196 Figure 8.5 : Simulated aortic valve flow (Qav), (black line) and pump flow ...... 198
Figure 8.6 : Simulated pump motor current (black) and reference aortic flow (Qav) (red) at fou r motor s-peed set points corresponding to (a) 50 rad/s, (b) 100 rad/s, (c) 150 rad/s and (d) 200 rad/s, where '■' and '●' indicate AV opening and closing times, respectively. Qav > 0 represents the periods when the AV is open, referred to as ventricular ejection (VE), and Qav =0 represents the valve closed (VC) state. Note that in panel (d), Qav = 0 at all times, indicating the AV remains closed throughout the cardiac cycle (ANO) for this pump set point speed of 200 rad/s...... 199
Figure 8.7 : Simulated aortic valve flow (Qav), (black line) and pump flow (Qp) (red line) at a pump set point of 50 rad/s...... 201 Figure 8.8 : Simulated left ventricular pressure (sold black line) and aortic pressure (dotted bl ack line) and pump differential pressure head (red line) at the low pump speed set point of 50 rad/s...... 202 Figure 8.9 :Snapshot of simulated LV blood velocity magnitude during LVAD support at various phases i.e. AV closure time (left figure in all panels), LV opening phase (two figures at the centre in all panels) and next cycle closure phase (right in all panels) during the cardiac cycle. (a) = 50 rad/s. (b) = 100 rad/s. (c) = 150 rad/s...... 203
x
List of Tables
Table 4 -1: Material properties used for the simplified LV model ...... 80 Table 4 -2: Mesh properties of simplified CVS LV-pump model...... 86 Table 4 -3: Material properties of the LV-LVAD interaction model incorporating CVS and heart contraction...... 90 Table 4 -4: Mesh properties of simplified LV-pump model with CVS and heart contraction. 92 Table 4 -5: Mesh properties of simplified LV-pump model without AV leaflets...... 98 Table 4 -6: Material properties of the myocardium in simplified LV model without AV ...... 99 Table 4 -7. Material properties of 2D realistic LV model ...... 107 Table 4 -8. Material properties of the myocardium ...... 107 Table 4 -9. Mesh properties of realistic LV model...... 109 Table 4 -10 Fluid material properties in 2D realistic LV model...... 115 Table 4 -11 Material properties for the solid model...... 119 Table 4 -12. Simplified LV-pump model parameters for the healthy and abnormal condition subjects: Systemic peripheral resistance, Rsa; and total blood volume, Vtotal...... 122 Table 4 -13: Realistic 2D LV-LVAD model parameters for healthy and abnormal conditions: Systemic peripheral resistance, Rsa; and total blood volume, Vtotal...... 123 Table 6 -1 A summary of physiological and pump basic parameters identifying AV state...... 148 Table 8 -1. A summary of physiological and pump basic parameters identifying the aortic valve conditions...... 193
xi
Part I-Introduction, Background and Literature Review
1
1 Chapter 1 - Introduction, Aims, and Outline
1.1 Research Motivation
Congestive heart failure (CHF) is a serious health condition resulting in diminished
blood flow to the tissues and organs in the body. More than 5 million people will
develop some degree of heart failure (HF) in their lifetime in the United States,
where its prevalence averages 2.1% of the normal population [1]. According to the
World Health Organization (WHO), a total of 17.3 million people died from
cardiovascular diseases (CVDs) in 2008, representing 30% of global deaths. Of
these, 7.3 million were due to coronary heart disease and 6.2 million were due to
stroke. This figure will increase to 23.3 million by 2030 [2].
Due to advancements in the medical treatment of HF, continuous flow Left
Ventricular Assist Devices (LVADs), known as Implantable Rotary Blood Pumps
(IRBPs), represent promising therapies for congestive HF patients [3]. A variety of
ventricular assist devices (VADs) have previously been effectively used in end-stage
heart failure patients, either as a bridge to heart transplant or to provide permanent
support. Patients alive on VAD therapy display significant improvements in heart
failure symptoms (New York Heart Association (NYHA) class II vs. IV), mood,
exercise capacity, and survival (1-year survival 52% vs. 25%; median survival 408
days vs. 150 days) [4].
2
More recent IRBPs are non-pulsatile blood pumps, representing the third generation of LVAD development. These non-pulsatile pumps are frequently used due their advantages of small size, efficiency, and reliability, which render them easily implantable, help improve patient health outcomes and increase the likelihood that patients resume normal lives [5]. Due to the shortage of appropriate heart donors,
LVADs are often used as permanent support, i.e. a destination therapy. They are also used as support to heart transplantation, i.e. bridge-to-transplant [6]. However, the increased use of LVAD for long-term mechanical support necessitates a better understanding of resulting hemodynamic changes in the left ventricle (LV), as well as the aortic valve (AV) state.
In addition, VADs can revise the systematic abnormalities in advanced CHF by improving systemic end-organ perfusion. Currently, VADs treat end-stage heart failure refractory to medical therapy [3]. A detailed understanding of the impact of the device on ventricular hemodynamic variables such as flow and pressure, as well as cardiac valve motion, is necessary to optimize LVAD use in unloading the ventricle.
On the other hand, in order to provide appropriate control strategies which accommodate heart valve state for the purpose of patient safety and recovery, an increased understanding of the interaction between the cardiovascular system (CVS), the LVAD, as well as the AV state under LVAD support, may allow the detection of the open/closed state of the AV from LVAD non-invasive parameters alone, paving
3
the way for more sophisticated pump flow control algorithms which take into
account the state of the aortic valve.
It is, therefore, important to understand the LV response to the pumping action of
these devices when connected to the heart, which can help improve the accuracy of
existing LVAD controllers which require feedback data such as blood flow and
pressure.
1.2 Thesis Aims
This thesis aims to examine the AV response to LV rotary blood pump assistance,
using 2D computational simulations of LV-pump interaction, investigating AV state
from LVAD non-invasive parameters. AV open-close state was assessed by
analysing pump motor current waveform (I) and pump impeller speed (ω).
The main hypothesis of this thesis was that AV state during LVAD support could
be detected and classified using non-invasive intrinsic pump parameters (i.e. pump
electrical current and pump impeller speed).
Using non-invasive pump parameters to investigate the impact of patient
physiological conditions on AV dynamics under different LVAD operating speeds,
4
may provide an insight to cardiovascular responses to these devices without using rotational examination (i.e. echocardiology). Furthermore, the AV is observed to be closed in most LVAD patients [7, 8]. However, with postural change, cardiac recovery and exercise, it may open on occasion. The AV may well offer a lower resistance path than the LVAD bypass route, with the native heart pumping significant flow relative to that via the LVAD [9]. These AV conditions under LV-
AD support could greatly influence greatly the current pump speed control strategy, and therefore would benefit from AV state detection.
AV state detection during LVAD support is challenging, and will greatly help to avoid high stress and pressure to prevent AV complications such as stenosis and thrombogenesis during LVAD support. Since implantable flow and pressure sensors are correlated with increase of the system cost and reduced reliability, this thesis will focus on using only non-invasive feedback measurements (i.e. motor electrical current and pump speed), as simulated from a 2D LV-LVAD interaction model, to develop all the necessary detection algorithms for AV state. Therefore, this thesis aims to achieve the following two objectives:
To develop a numerical 2D LV-LVAD model using Finite Element (FE), Fl-
uid Structure Interaction (FSI) methods, to enhance the development of AV
state (open/close) detection methods during LVAD support. The model was
formulated with the aim of providing insights into the dynamics of heart-
pump interaction that cannot readily be obtained experimentally.
5
To develop a novel approach to detect and monitor AV function during LV-
AD support by using simulated LVAD motor current waveform as an index
signal. Such information will lead to a more precise evaluation of AV functi-
on, helping reduce the risk of aortic insufficiency and other AV problems.
This AV dynamics data could be used in developing a LVAD feedback
controller with the objective of recognizing pump speed set points within a
safe range of pump operating speeds, where the LVAD could be driven at a
pump speed between the starting point of 1) partial assistance (i.e. pump
assistance ratio < 100% where the net positive flow through both the aortic
valve and the LVAD) and 2) total assistance (i.e. pump assistance ratio =
100% where the AV remains closed throughout the entire cardiac cycle).
1.3 Thesis outline
This thesis is presented in nine chapters describing some background of HF, possible
treatments, modelling methods used, and simulation results of LV-LVAD fluid-
structure interaction in two-dimensional models that mimic the status of the aortic
valve during LVAD support.
Chapter 2 of this thesis introduces basic cardiovascular physiology, including AV
function and a brief discussion of HF and AV abnormal pathophysiology in
6
LVAD patients. Methods for HF treatment are briefly explained, focussing on LVADs and their various types (i.e. continuous blood flow and pulsatile flow) an d their applications. A summary of computational FE, FSI methods are also included in this chapter, as well as some details on existing LVAD control systems with methods of estimating pump parameters.
Chapter 3 continues with a review of the literature into this emerging field of FSI
LVAD modelling, allowing the reader to understand accumulated work done to resolve the problem of AV state detection and function. A survey will be presented regarding LV modelling, examining Computational Fluid Dynamics (CFD) approaches which have been used to model the cardiac cycle including LV and AV dynamics. Various studies that have attempted to simulate AV function under
LVAD support will be also be discussed.
Chapter 4 defines the computational methods used in this thesis, outlining two major geometrical LV-pump models. The first describes a modelling approach based on highly-simplified LV-Pump geometries, including dimensional definitions and numerical settings. The second model describes a geometric approach based on a more-realistic LV-Pump anatomy, including dimensional definitions and numerical settings. These models are used for various purposes:
(i) to simulate and predict AV status during LVAD support.
(ii) to examine the resulting data for pump current, average aortic outflow and distance between leaflets waveforms, under pulsatile flow conditions.
(iii) to validate the simulated models.
7
(iv) to estimate average pump differential pressure and average pump flow rate fr- om average pump input power.
Lastly, in this chapter I have underlined the methods and protocols of simulation used to investigate LV-pump interaction over a wide range of pump and physiological operating conditions, including variations in:
(i) systemic vascular resistance; and
(ii) total blood volume, by varying the inlet-flow rate to the LV.
The primary focus of Chapters 5 and 6 is to use the 2D LV-pump model to simulate the AV open-closed state, and determine the effect of intra-ventricular flow on AV state within the LV chamber under LVAD support. Chapter 6 was divided into two sections: the first presents a systematic approach using a simplified LV-Pump interaction model to predict the AV state during LVAD support, whilst the second describes a more-realistic LV-Pump geometry with AV, to predict AV dynamics during LVAD support.
Chapter 7 represents an extension to Chapter 6. Three pump speed set points were used to investigate LV-pump interaction over a wide range of pump operating condi- tions and physiological states, including normal heart condition as well as alterations in systemic vascular resistance and total blood volume.
Chapter 8 represents an extension to Chapter 6, in which the aortic valve model was replaced with a computational representation of a pressure valve to mimic aortic 8
outflow. The AV state was predicted during LVAD support by using aortic flow and
non-invasive pump parameters.
Finally, recommendations for future research, along with the main conclusions of
this thesis are presented in chapter 9.
1.4 Publications
The work presented in this thesis was published in the following refereed conference
proceedings and abstracts during my PhD candidacy:
1.4.1 Refereed Conference Proceedings and Abstracts
K. A. Alonazi, A. V. Savkin, N. H. Lovell, and S. Dokos, ".Use of an implantable rotary blood pump for sensorless estimation of ECG isoelectric potential in a model of heart failure," in Engineering in Medicine and Biology Society (EMBC), 2012 34th Annual International Conference of the IEEE, 2012.
K. A. Alonazi, N. H. Lovell, and S. Dokos, "Modelling aortic valve open-close events during left ventricular support," Australian Biomedical Engineering Conference (ABEC), 2013.
K. A. Alonazi, N. H. Lovell, and S. Dokos, "Simulating assist device motor current waveform in monitoring aortic valve state during left ventricular support," Australian Biomedical Engineering Conference (ABEC), 2014.
K. A. Alonazi, A. V. Savkin, N. H. Lovell, and S. Dokos, "Modelling aortic valve closure under the action of a ventricular assist device," in Engineering in Medicine and 9
Biology Society (EMBC), 2013 35th Annual International Conference of the IEEE, 2013, pp. 679-682.
K. A. Alonazi, N. H. Lovell, and S. Dokos, "Simulation of motor current waveform as an index for aortic valve open-close condition during ventricular support," in Engineering in Medicine and Biology Society (EMBC), 2014 36th Annual International Conference of the IEEE, 2014, pp. 3013-3016.
K. A. Alonazi, N. H. Lovell, and S. Dokos, "Simulation of motor current waveforms in monitoring aortic valve state during ventricular assist device support," in Engineering in Medicine and Biology Society (EMBC), 2014 IEEE Conference on Biomedical Engineering and Sciences (IECBES), 2014.
10
2 Chapter 2 - Background
2.1 Overview
This chapter provides a brief background to the research area of this thesis. Basic
cardiovascular anatomy and physiology related to both the heart and the aortic valve
are presented in sections 2.2, followed by an overview of congestive heart failure and
its treatments in section 2.3, since this thesis aims to detect and monitor AV function
during LVAD support in heart failure. This section also includes a brief overview of
various pump outflow control strategies currently adopted (section 2.3.4), since this
thesis aims to provide insights into LVAD control based on simulated pump motor
current to monitor AV function. Section 2.4 then presents an overview of existing
heart-pump interaction computational techniques and models. More detailed literature
reviews are given in subsequent chapters in order to complement the distinct topics
presented.
2.2 Basic Anatomy and Physiology
2.2.1 Cardiovascular system
The human heart is located in the mediastinum (the space between the lungs in the
thoracic cavity), and weighs approximately 300 grams. The heart incorporates the
right and left atrium, the right and left ventricle (LV) and valves. The atria pass the
blood into the ventricular chambers. In contrast, the right and the left ventricles have
11
a thicker muscular wall to generate significant force to pump the blood through either the pulmonary or the systemic circulation [10] (see Figure 2.1).
The valves are located at the outlet and inlet of both ventricles to prevent the blood from backflow. For example, the mitral and tricuspid valves, during systole, prevent blood from returning to the atria from the ventricles. Likewise, the pulmonary and aortic valves prevent backflow to the ventricles from the aorta and pulmonary arteries during diastole.
Figure 2 .1:A longitudinal view of the heart and its main components (adapted from [10]) .
The cardiovascular system is comprised of both the heart and circulation, and functions to distribute essential elements to the tissues and organs to regulate and adjust oxygen supply and nutrients. It includes the heart, blood vessels, as well as major arteries, and veins which carry the blood (i.e. oxygenated and deoxygenated) 12
from the heart to the body tissues and back to the heart through the systemic and pulmonary systems [11] (see Figure 2 .2). The blood circulates through a closed system starting from the heart, passing through arteries, capillaries and veins. The systemic circulation covers the greater proportion of the blood in the circulation
(84%).
Figure 2 .2: Distribution of blood volume in the different parts of the circulatory system, (adapted from John et al. [11]).
The systemic circulation originates from the LV through the aorta, an elastic artery approximately 2.5cm in diameter, passing through various systemic tissues via the large arteries, small arteries, capillaries, small veins, and large veins before flowing back to the heart into the right atrium and right ventricle (RV). Blood return to the right atrium occurs via the superior and inferior venae cavae. Then RV then pumps the deoxygenated blood to the lung, through the pulmonary arteries, for gas exchange. The blood is then returned to the left atrium through the pulmonary veins
13
(see Figure 2.1), flowing to the LV to be pumped again into the systemic circulation as shown in Figure 2 .2 [11]
Figure 2 .3: “Syncytial” interconnecting nature of cardiac muscle fibers (adapted from Guyton et al. [12]).
Heart muscle is constructed of cells known as myocytes, connected together via intercalated discs, formed by cell membranes, and specific for cardiac muscle [12].
Cardiac muscle is made from bundles of myofibrils, which contain the contractile elements (see Figure 2 .3).
Figure 2 .4: Ventricular pumping action; systole and diastole (adapted from Mohrman et al. [13]).
Heart muscle contracts by electrical stimulation from impulses arising from excitation of neigbouring cells. This excitation (known as an action potential) travels 14
from cell to cell. As a result, free diffusion of ions occurs and the action potential
spreads from one cell to another if these cells are excited.
The cardiac cycle begins from action potentials which originate in the sinoatrial
(SA) node located in the right atrial wall. These travel through both left and right
Figure 2 .5: Left ventricular function during cardiac cycle illustrating variations in left atrial pressure, left ventricular pressure, aortic pressure, pressure gradient, the electrocardiogram (ECG), and the phonocardiogram (adapted from Yellin et al. [14]).
atria causing contraction. The action potential is then conducted through the atrio-
ventricular bundle into the venricles. During the contraction period, blood is pumped
from the left ventricle through the aorta to the rest of the body, and this phase is
called systole [13] (see Figure 2 .4). Diastole is the period when the ventricles relax,
in order that they refill with blood returning via the venae cavae. Slow passive filling
of the ventricles marks this initial diastole phase (diastasis), with the final phase of
ventricular diastole known as atrial systole, where the atria contract with an ‘atrial 15
kick’, to fill the ventricles with 20% of their blood capacity. As shown in Figure 2 .5,
once the blood is transferred to the ventricles, they then contract, with this
contraction revealed as QRS waves on the ECG trace about 0.16 second after the
onset of the P wave [14]. Ventricular pressure then rises, causing both the mitral and
tricuspid valves to close. After this period when both valves are closed, the
isovolumic phase of contraction begins, where the pressure continues to build up
inside the ventricle.
When ventricular pressure rises above that of the pulmonary trunk or aortic pressure
at the end of isovolumic contraction, the aortic and pulmonary valves open, forcing
blood flow from the ventricles. Following this phase of ejection, the heart undergoes
a period of ventricular repolarization and relaxation, evidenced as a T wave on the
ECG [14] (see Figure 2 .5). As a result, ventricular relaxation starts and ventricular
pressure drops below that of the pulmonary arteries or aorta. The blood flows back
to the ventricles, resulting in both aortic and pulmonary valve closure, holding the
blood volume inside the ventricular chambers fixed. The ventricular pressure falls
rapidly below the atrial pressure causing the mitral and tricuspid valves to open,
allowing the blood to pass to the ventricles again, starting a new cycle of cardiac
contraction and relaxation.
2.2.2 Control of Cardiac Output
The heart in its natural environment pumps more than 7 tons of blood daily, beating
more than 100,000 times per day to meet the perfusion demands of the tissues and 16
organs. Due to this complex pumping action to maintain the pulmonary and systemic circulations, the heart requires a significant control system to guarantee accurate, continuous, and constant regulation of its haemodynamic parameters (i.e. pressure and heart rate), adjusting in response to external factors like stress, exercise or haemorrhage [13]. Heart function can also be described by the mean left ventricular pressure and volume of blood pumped per unit time (i.e. cardiac output). Heart pumping power is altered in HF patients, and can be severely diminished to the extent that the heart is unable to supply enough blood to meet the body’s metabolic needs.
Figure 2 .6: Heart pump function. The gray arrow indicates the influence of increased ventricular filling, while the black arrows indicate the findings of Frank and Starling ( adapted from Vandenberghe [15]).
The heart operates and is intrinsically controlled according to Starling’s law of the heart, known as the Frank-Starling mechanism which states that, Stroke volume increases as the volume of blood returning to the ventricle (end-diastolic volume
(EDV) or preload) increases [13] (see Figure 2 .6 ). This can be explained by the intrinsic mechanical properties of heart muscle: the greater a heart fibre is stretched, 17
the greater the extent of overlap between thick (myosin) and thin (actin) filaments,
leading to the formation of more cross-bridges and an increase in the fibre’s
contractile force. In other words, for a given afterload, increasing the preload to the
heart increases the active tension produced by the heart muscle fibres.
However, another approach for measuring cardiac function is via the LV pressure-
volume relationship. As illustrated in Figure 2.7 the LV pressure volume loop indi-
cates four phases that are representative of ventricular function: isovolumic relaxati-
on, filling, isovolumic contraction, and ejection [15].The relation between preload
(filling pressure) and end-diastolic volume has vital physiological and clinical
consequences. For example, in a normally functioning cardiac muscle, the actual
relationship is curvilinear, especially at very high filling pressures; it is
linear over normal operating range of the heart [16]. However, HF is characterized
by an abnormal pressure-volume relationship.
Figure 2 .7: Typical pressure-volume relationship of the left ventricle. The letters indicate valve action: A. mitral valve (MV) opens; B. mitral valve closes; C. aortic valve opens; D. aortic valve closes. (adapted from Vandenberghe [15]).
18
Other factors that have an effect in controlling the function of the heart include the
systemic arterial pressure (AP) effect on end-systolic volume during heart failure.
Furthermore, cardiac sympathetic and parasympathetic nerve activity also affects
contractility of the heart as well as its rate of contraction [17].
2.2.3 Aortic Valve Anatomy and Physiology
The heart has four valves that control the direction of blood flow through the heart.
Heart valves play a crucial role in regulating the flow of blood between chambers.
Their function is to control the direction of blood flow during the cardiac cycle, or in
other words, to prevent the backflow of blood. Essentially, heart valves are passive
tissues that open and close due to inertial forces exerted by the surrounding blood.
There are two types of valves: semilunar and sigmoidal (see Figure 2.1). The
semilunar valves, (pulmonary valve (PV) and aortic valve (AV)), control the flow of
oxygenated blood from the left side of the heart to the body, and are open during
systole when the ventricles are contracting to prevent the blood from flowing back
into the ventricle during diastole [10]. During isovolumic contraction and relaxation,
both valves are closed. These types of valves have different size and leaflet
dimensions.
19
Figure 2 .8: Schematic representation of the aortic valve: (a) side view of the complete valve, (b) after dissection of one leaflet with corresponding sinus wall and, (c) aortic view, (adapted from De Hart et al. [18]).
On the other hand, the aortic valve acts as a one way valve between the left ventricle
and the ascending thoracic aorta, as shown in Figure 2 .8. Basically, the valve
consists of three membranous leaflets and three sinuses [18]. These structures form a
cylindrical wall to which three crescent-shaped leaflets are attached. These leaflets
are attached to the aorta, and with their free edges, they fold back towards the wall
of the aorta to permit blood to flow from the left ventricle during systole. The
leaflets come together during diastole to prevent backflow and assist in the efficient
LV filling with blood.
As shown in Figure 2 .8, the cusp adjacent to the aortic valve is designated as the
anterior or aortic cusp, while the one closer to the ventricular wall is designated as
the posterior or mural cusp. Behind the cusps, the aortic root forms bulges known as
the sinuses of Valsalva, which, as reported by Bellhouse et al. [19], play a role in the
closure mechanism of the valve. They intersect the aortic root wall on the ventricular
side in a crown-like shape. The upper level where the sinuses merge into the aorta is
20
known as the sinotubular junction, and corresponds to the upper reach of the leaflets
in the open position.
2.2.4 Regulation of the Aortic Valve
During systole, which normally lasts 200 to 300 ms [20], the LV contracts and the
aortic valve opens. This valve closes during diastole when the LV is filling through
the open mitral valve. During isovolumic contraction, both valves are closed [21]
(Figure 2 .9). If the heart beats at 65 times per minute, which is a normal heart rate at
rest, then a whole heart cycle would last 0.95 seconds. The duration of ventricular
diastole would be almost two thirds of the beat, while systole would last about one
third of the cardiac cycle and begin when the aortic valve opens. During systole,
blood quickly rushes through the valve and reaches a peak velocity during the first
third of systole after the leaflets have opened to their full extent and are beginning to
close again. The flow starts to slow rapidly after the peak is reached. The developed
adverse pressure gradient affects the low momentum fluid near the wall of the aorta
more than that at the centre: this causes reverse flow in the sinus regions [21].
21
Figure 2 .9: Pressure and flow curves for aortic and mitral valves (adapted from Yoganathan et al. [21]).
The pressure difference between the ventricle and aorta determines blood flow
through the aortic valve. Figure 2 .10 illustrates the pressure and flow relations across
the aortic valve during the cardiac cycle [22]. During systole, the pressure difference
required to drive the blood through the aortic valve is small; however, the diastolic
pressure difference reaches 80 mmHg in normal individuals. The aortic valve closes
near the end of systole with very slight reverse flow through the valve.
22
Figure 2 .10: Aortic (Pao: dashed line) and left ventricular (Plv: solid line) pressure curves during the cardiac cycle. The associated flow curve is also given (Q: dotted line). AO denotes the onset of valve opening and AC the moment of complete closure (adapted from Yoganathan et al. [22]).
Furthermore, the heamodynamics of the heart are responsible for shaping aortic
valve functionality, and knowing these effects are essential to understand the heart
response under LVAD support. For example, in healthy individuals, blood flows
through the aortic valve and may reach a peak velocity of 1.35±0.35m/s [23,
24]. Blood flow through the aortic valve is affected by the state of the valve.
Therefore, the complete closure and opening of the valve during and at the end of
systole is controlled by the inertial flow produced from the pressure difference
developed in the boundary layer along the aortic valve wall.
In addition, development of vortices in all the three sinuses behind the AV leaflets,
have a function was first described by Leonardo da Vinci in 1513. With help of these
vortices during systole in the sinuses, the leaflets are pushed away from the wall to
close the valve, and prevent blood from returning to the ventricle during the closing
process, assisting in the efficient closure of the valve [25].
23
There are various studies that have investigated the action of vortices on AV closure
[26, 27], including the role of transverse pressure differences created by vortices that
push the leaflets toward the centre of the aorta and each other at the end of systole.
They concluded that this transverse pressure difference alone is enough to close the
valve, but is not as efficient and rapid in the presence of vortices. The next section
will provide a brief overview of the of congestive heart failure and its treatments,
since this thesis aims to detect and monitor AV function during LVAD support in
heart failure.
2.3 Heart Failure Pathology and Therapy
2.3.1 Congestive Heart Failure
Congestive Heart Failure (CHF), or Heart failure (HF), is characterized as a serious
condition caused when cardiac pumping function reduces to such an extent that the heart is
unable to supply sufficient blood to cope with the body's physiological needs. The resulting
weak pumping action of the heart leads to a build-up a fluid (congestion) in the lungs and
other body tissues including the legs and feet. According to the World Health Organization,
the global incidence of CHF is increasing each year, with an estimated 9.4 million deaths
each year caused by heart disease [28]. CHF always results from an underlying disease
primarily affecting heart function such as heart valve disease or cardiomyopathy. It occurs
when the heart is unable to deliver sufficient cardiac output (CO) to prevent venous
congestion. When blood is not adequately returned to the heart, increasing pressure in the
lung and the veins leads to a build-up of fluid in different parts of the body. 24
Moreover, Gary et al. [29] describes heart failure as a “complex syndrome that can result from any structural or functional cardiac disorder that impairs the ability of the heart to function as a pump to support a physiological circulation”. Furthermore, the severity of CHF is difficult to quantify as it has various potential causes: it may arise from a number of defects such as myocardial failure where the cause is due to damaged myocardium (structural abnormalities), or mechanical factors where some obstruction to the pumping action of the myocardium leads to a decrease in its effectiveness, or from a combination of both the above. Increased understanding of the causes of CHF will lead to better therapeutical approaches. Patients suffering from end-stage CHF due to a damaged heart such as ventricular hypertrophy or dilatation are mainly treated with cardiac assist devices.
Indices used to predict cardiac function include ejection fraction, end-systolic volume (ESV), mean velocity of circumferential fibre shortening, end-systolic pressure-volume relationship, end-systolic stiffness, preload recruitable stroke work and first time-derivative of ventricular pressure [30]. In addition, increased pumping power of the heart in a CHF patient is accompanied by a reduction in stroke volume and cardiac output, as well as an increase in end-diastolic pressure (EDP), end- systolic volume and end-diastolic volume (EDV) of the failing ventricle [31, 32]. In the presence of reduced cardiac function, a number of natural body mechanisms occur to manage cardiac output including reflex control mechanisms which improve
abnormal contractility, left and right heart remodelling, as well as fluid retention
[31, 33, 34].
25
Figure 2 .11: Cardiovascular alterations with compensated systolic heart failure. Point (A) illustrates the intersection of normal CO and normal venous function curves (adapted from Alomari et.al [13]).
During HF, regardless of the cause, the pumping power of the heart and the Frank-
Starling mechanism are both degraded. The conditions of chronic heart failure are illustrated by the CO and venous function curves, as shown in Figure 2 .11 Alomari et.al [13]. These curves illustrate the normal operating point of the heart, defined as the intersection between the normal CO and normal venous function curves. In the healthy heart state, CO is 5 L/min and venous pressure almost 2 mmHg, whereas in the HF condition, the CO will be lower than normal at any given filling pressure causing a shift from the normal operating point as shown in Figure 2 .11. In addition, for the HF condition, the decline in SV and CO will cause an increase in all heart dynamics parameters such as contractility, total blood volume, and myocardial
26
remodelling, due to the increase in sympathetic neural activity, coupled with the increase in end-diastolic pressure as well as increased end-systolic and end-diastolic volumes [34, 35].
Due to HF, the relationship between ventricular pressure and volume is affected by a decrement in stroke volume and ejection fraction. Figure 2 .12 Alomari et.al [13] shows these phenomena along with a high filling pressure, marked by a decrease in contractility of the heart, as evidenced by the downward shifting of the slope of the end-systolic pressure volume relationship.
Figure 2 .12: Left ventricular pressure-volume loops during heart failure ( adapted from Alomari et. [13]).
Due to the complexity of CHF and its different possible causes and symptoms, the severity of HF is commonly gauged using the New York Heart Association (NYHA)
Functional Classification method based on the function condition of the patient. It
27
provides a way of classifying the extent of HF in patients according to their
symptoms [36]. These are:
I. Patients with no limitation in physical activities; they suffer no symptoms when
undertaking such activities.
II. Patients with slight, mild limitation during physical activity; they are comfortable
with rest or with mild physical exertion.
III. Patients with marked limitation of physical activity; they are comfortable only at
rest.
IV. Patents unable to carry out any physical activity without discomfort; they should be
at complete rest, confined to a bed or chair.
2.3.2 Ventricular Assist Devices: an Emerging Therapy for CHF
Therapeutic options for treating end-stage heart failure have evolved significantly in
recent years. Transplantation of the heart remains the traditional treatment, and is the
only therapy offering a real solution for the chronic HF patient. The ratio of the
number of donor hearts available to the number of potential recipients is declining.
This due to the fact that CHF is the main cause of death in comparison with
other diseases [15]. Due to the lack of donor organs, various alternative therapies for
CHF have emerged in the healthcare system. These include medical therapies such
as pharmacological treatment and surgery. However, mechanical therapies such as
total artificial heart replacement and ventricular assist devices (mechanical 28
circulatory support) are alternative treatments for heart failure. This thesis will focus
on left ventricular assist devices.
2.3.2.1 Surgical Therapies for Heart Failure
2.3.2.1.1 Heart transplantation
Despite advances in new technologies for the treatment of severe HF, the numbers of
myocardial failure patients have failed to drop considerably. The American Heart
Association’s latest update estimates that nearly 5,000,000 Americans suffer from
heart failure. Nearly 10 % of the population over 75 years old have been diagnosed
with HF [37].
Heart transplantation has become the standard therapy for these patients, and is
commonly accepted to reduce mortality and morbidity, as well as improve quality of
life. However, heart transplantation is a major surgical procedure carried out on end-
stage HF or severe coronary artery disease patients. Nonetheless, it is still the most
reliable solution for patients who have reached end-stage HF.
29
Figure 2 .13: Prof. Christiaan Barnard with the first recipient of a donor heart (adapted from Vandenberghe [15]).
The first successful human heart transplant was performed by Christian Barnard in
1967 in the Groot Schuur Hospital in South Africa (Figure 2 .13 Vandenberghe [15]),
followed by Shumway in 1968 in the United States at Stanford. Currently, more than
5000 heart transplant procedures are performed annually worldwide [29]. Although
heart tran-splantation still remains the standard treatment for heart failure, as
reported by Hardy et al. [38], due to the limited number of donors and graft failure,
multi-organ-system failure, infection, and acute rejection are the major leading
causes of death following transplation, minimizing the applicability of this
treatment. However, alternative medical therapies are increasingly being recognized
through medical trails as options for restoring heart function, alleviating symptoms,
and improving clinical outcomes.
30
2.3.2.1.2 Surgical reshaping
HF can also be treated with a number of innovative surgical approaches, which aim
to improve ventricular function, quality of life, and eventually, survival. These
approaches include coronary artery bypass surgery, aortic valve replacement, mitral
valve repair, ventricular restoration and passive restraints. In the following section,
two surgical treatments will be elaborated briefly.
2.3.2.1.2.1 Ventricular volume reduction Ventricular restoration, also known as the “Batista procedure”, is a treatment for
dilated cardiomyopathy which focuses on the reduction of ventricular volume to
restore ventricular function, by removing a wedge shaped piece of the LV wall of
about 100 g [34]. (see Figure 2 .14 Starling et al. [38]). In 1994, this treatment was
first implemented by Randas Batasta as a new surgical technique called partial left
ventriculectomy, He developed this technique because he lacked the technology that
was available in Europe and North America to use other therapies. His technique
was used to correct the pathologic alteration in cardiac geometry from its abnormal
spherical shape, caused by infraction, to its normal elliptical shape. Furthermore, this
procedure was performed by removing a wedge piece from the left ventricular wall.
However, in 2004 Dor [39] modified this technique, resulting in improvements in
the functional limitations of patients with heart failure end-systolic volume index
(ESVI) was calculated as end-systolic volume (ESV) divided by body surface area
31
[40]. Hence, this procedure can be used with transplant candidates, so it is also
applicable to carry out heart transplantation.
Figure 2 .14: Schematic of the myocardial wedge excised from the left ventricle in the Batista procedure. The resulting exposed edges of the ventricle are subsequently sewn together to yield a ‘remodelled’ ventricle (adapted from Starling et al. [38]).
2.3.2.1.2.2 Dynamic Cardiomyoplasty Another important surgical therapy for HF is dynamic cardiomyoplasty or passive
restraints. This procedure is based on hemodynamic factors to improve the structural
deterioration linked with heart failure. It is performed by wrapping the left latissmus
dorsi skeletal muscle around the dilated heart and stimulating it to contract
synchronously with the heartbeat by applying a regiment of electrical pulses. The
key issue in this concept is the training and stimulation of the muscle.
Cardiomyoplasty, as stated by Stijn [41], was first used by Carpentier following
testing with animal experiments. Today, this procedure is sufficient in patients with
32
congestive heart failure, by keeping the heart dimensions stable over a long period,
thus eliminating further dilation and remodelling of the heart [38].
2.3.2.2 Pharmacological therapy
Drugs have always been part of HF treatment, since they unload the heart by
controlling its activity or decreasing the afterload pressure from the arterial system.
However, the goals of pharmacological treatment are to minimize the likelihood of
death, alleviate symptoms, and improve the quality of life. These are achieved
through the use of inotropic agents such as digitalis, which increase the power of
contraction of the heart by increasing the amount of calcium ( released in the
myocytes. However, it is only in the early stage of heart failure that such drugs are
utilized [42]. Furthermore, β-Blockers and angiotensin-converting enzyme (ACE)
inhibitors are also very popular drugs for the treatment of HF. ACE is an enzyme
that is produced in the kidney, responsible for maintaining the tone in the blood
vessels as well as relaxation of the vessels and reduction in afterload. Unfortunately,
some of these drugs have unknown biochemical mechanisms, and can have toxic
side effects which result in excessive lowering of blood pressure to undesirable
levels [31]. However these treatments are also limited in their effectiveness, with an
associated increase in mortality by almost 18% every 6 months [29]. In addition,
these types of drugs may initially worsen ventricular haemodynamics, including the
β-blockers which affect heart rate and contractile power of the heart, decreasing
oxygen demand.
33
2.3.2.3 Mechanical therapies
Physicians nominated the option of replacing a defected heart with an artificial one,
even before even heart transplantation was available. The first successful total
artificial heart (TAH) was developed in 1953 by Gibbon, who developed the first
heart-lung machine during a cardiopulmonary bypass (CPB) surgical operation for
45 min [38]. However, due to the complications linked to these devices, total
artificial hearts are still being researched as a viable mechanical option for HF.
Currently, there are many disadvantages and challenges to deal with before any
TAH is globally accepted and used on patients, including life-threatening infection,
bleeding and thromboembolism [36] .
Another mechanical treatment for end-stage HF patients is the use of a cardiac assist
device (CAD) also referred to as a ventricular assist device (VAD). VADs are now
accepted as alternative treatment therapies for congestive HF patients. Because the
number of donors are limited, VADs are often essential, and are used to assist the
failing heart by pumping blood from the left or the right atrium or ventricle through
the ventricular assist device to a major artery [43]. In some cases, they can be used
to support both ventricles, left and right, also known as bi-ventricular assist devices
(BVADs or BiVADs)
Ventricular assist devices were first used in 1961 to support a patient whose right
ventricle exhibited impaired pumping [44]. The first implantable left VAD was rep-
34
orted in 1963 by Liotta et al. [7]. Furthermore, VADs can be implanted as a bridge to heart transplantation, or even as a destination therapy. They can improve end- organ function and reduce morbidity. To this end, pulsatile and non-pulsatile pumps have been developed to provide support of the natural heart. Eventually, it is desirable to establish a controller for such blood pumps, utilizing information such as blood pressure and pump flow rate. These two types of pumps are illustrated in
Figure 2 .15 [5]. Implanted LVADs draw blood from the apex of the left ventricle and pump it to the ascending aorta. Both pulsatile and non-pulsatile types are electrically energized by means of a percutaneous lead that connects the pump to an external system controller and power source.
DeBakey implanted the first ventricular assist device in 1963, to successfully treat a patient who had suffered a cardiac arrest following an aortic valve replacement.
Although the patient subsequently died, this operation proved the feasibility of mechanical circulatory assistance [37]. The limitations of these early devices, however, provided the motivation for the development of an intracorporeal LVAD powered from an external power supply. Currently, with the development of these devices, VADs are being successfully used as a bridge to recovery. This means the device does not only prolong life, but also improves the quality of life for HF patents with short life expectancy. For example, the Randomized Evaluation of Mechanical
Assistance for the Treatment of Congestive Heart Failure trial (REMATCH) was preformed over 3 years with over 120 patients being under observation with implanted ventricular assist devices. The outcomes established survival rates with a significant reduction of 33% in mortality, as well as significantly higher quality-of-
35
life benchmarks [45]. Furthermore, the expansion of the VAD spectrum is clearly evident today, and various studies have shown that LVADs can be successfully used as a bridge to transplantation. For example, a study designed for the second- generation LVADs showed successful outcomes with patients being assigned to ventricular assist devices, with survival rates 70% in the second year post- implantation [5, 44]. Consequently, these patients were able to return to normal daily activities. Despite the significant developments in LVAD technology, major problems such as thrombosis, bleeding and LVAD failure have not yet been solved, indicating much more room for LVAD improvement.
The LVAD results in a decrease in patient morbidity at the time of transplant, rendering the patient a better candidate and improving his/her post-transplant outcomes. The LVAD takes over the work of the LV by decreasing the filling pressure and increasing CO. Clearly, it will increase blood perfusion in organ systems, allowing HF patients to stabilize as they wait for a heart transplant. Another study in 10 patients who received an LVAD as a bridge to transplantation was maintained for up to 214 days. It found that LVADs maintain renal and hepatic function throughout the period of circulatory assistance [46].
According to the pumping principle of VADs, two major types can be classified: pulsatile-flow pumps and continuous-flow pumps. It is not the intention here to provide an overview of all the available devices. However, a short summary of current devices along with some new types will be given below.
36
Figure 2 .15: Pulsatile-Flow (Panel A) and continuous-Flow (Panel B) Left Ventricular Assist Devices (LVADs) (adapted from Slaughter et al. [5]).
2.3.2.3.1 Pulsatile-Flow ventricular assist devices
The Pulsatile-Flow VAD has been used medically since the late 1970s. It
consists of a chamber that fills and empties intermittently. In other words,
these devices have a chamber that fills with blood during one stage of the
operation cycle and pumps out the blood during the subsequent phase of the
cycle. This type of assist device may be divided into centrifugal, axial and
diagonal rotary pumps that differ in their appearance and inflow blood path.
37
In the device shown in Figure 2.15 (Panel A) Slaughter et al. [5], the motor withi n the device is powered from an external supply box. The motor drives a pusher
plate up and down frequently, that presses on a polyurethane diaphragm, whic h will maxim-ize and minimize the chamber’s volume and deliver a pulsatile flow to the aorta. In addition, it has a built-in valves at the inflow and outflow to control the direction of blood flow.
Furthermore, the TAH may also be considered a type of pulsatile-flow pump, where the patient’s ventricles are removed and the pump is inserted orthotopically [47]. Another example of this device is the third generation of the impeller rotary pumps with free contacting components and virtual shafts
Figure 2 .16: VentrAssistTM pump ( adapted from Gosline [17]).
38
supported by magnetic bearings, such as the VentrAssistTM pump (Ventracor Li-
mited, Sydney, Australia). In these devices, hydrodynamic forces offer a simple
means of suspending the impeller, preventing the complexity of active magnetic
bearings and reducing wear of contact pivot bearings. Consequently, these
devices are small, light, resistant to infection and durable, which makes this
generation of IRBPs easily to implant. Subsequently, good performance has
been achieved in device efficiency and prevention of haemolysis [8, 16, 48]. The
limitations of this type of LVAD are that the pump is noisy and the valve inside
the pump will decay over time. In this thesis, the LVAD that will be modelled is
the VentrAssistTM device, shown in Figure 2 .16 [17].
2.3.2.3.2 Continuous-Flow ventricular assist devices
As the blood moves through the systemic circulation, the initial pulsatile flow in the
aorta is gradually reduced, becoming continuous flow at the level of the capillaries.
Accordingly, pulsatile flow may not be necessary for humans. The non-pulsatile-
flow pump shown in Figure 2 .15 (Panel B) Slaughter et al. [5] contains a titanium-
coated rotor that curves around a central shaft with helical blades and an impeller.
An external drive line provides electrical power to a motor that drives the rotation of
the impeller by electromagnetic induction. The spinning impeller draws blood from
the inflow cannula to the outflow cannula. With this design, the continuous-flow
device has only one moving part and does not require valves, minimizing the long-
term risk of mechanical failure. In addition, the pump is smaller and lighter than the
first generation LVADs and functions in virtual silence.
39
Early concerns about the adverse effects of continuous flow on end organs have
been alleviated with successful long term use of this device [3]. Currently, the
HeartMateTM II is the only Food and Drug Administration (FDA)-approved
continuous flow pump.
2.3.3 Abnormal pathophysiology in LVAD patients
2.3.3.1 LVAD-Related Aortic Valve Dysfunction
In the past decade, studies have shown a potential correlation between LVAD
support and heart valve dysfunctions, in both pulsatile and non-pulsatile pump types
[49-54]. However, many studies did find a link between AV complications and
LVAD use. For example, one study found evidence of AV commissural fusion (i.e. a
remodelling process in which the aortic valve leaflets adhere together to prevent
complete opening of the valve) explanted from hearts supported by an LVAD [8,
51].
2.3.4 LVAD control systems
The centrifugal blood pump LVAD has been developed in many research
laboratories world-wide to understand its impact on the CVS with the aim of
developing a robust, implantable mechanical support device. Efforts have been made
to study and develop a variety of LVAD control systems in order to achieve the 40
optimal pump output demanded by the patient’s circulatory system. However, many
research groups have agreed that the main purpose of these control systems is to
provide a long-term treatment for heart failure. On the other hand, the implantation
of additional sensors is not desirable, as they may result in thrombus formation and
require regular calibration due to measurement drifts, which makes long-term
implantation of such devices problematic.
2.3.4.1 Dynamic modelling and estimation of LVAD pump parameters
The LVAD drives blood through the circulation by supporting the function of the
heart. In fact, it mimics the natural heart function. According to the Frank-Starling
law, the output of the ventricles depends upon the preload in such a way as to
provide blood flow that is equal to venous return. However, most researchers have
aimed at simply representing the function of the heart by monitoring LVAD
performance. One approach is to apply the intrinsic characteristics of the pump to
estimate the hemodynamic state of the patient. The second approach directly moni-
itors pressure and flow through the pump.
Blood pump operation requires the blood pressure and flow as parameters to the
controller, which can either be directly measured or predicted by an estimation
technique. Funakubo et al. proposed a method of estimating rotary pump flow rate
and pressure head from its motor speed and torque [55]. This approach was to be
used with a conventional positive-displacement type pump, where differential
41
pressure is estimated from the power supplied to the pump [56]. However, this type of pump failed when pump speed set point was increased because of back-flow, resulting in negative pressure or suction. More extensive control is obviously needed, not only for the pumping action of the VAD, but also considering the factors which the characteristic curve of the pump is highly sensitive to. These factors include physiological and mechanical properties such as blood viscosity and impeller inertia of the pump. Therefore, another scheme has been proposed to control the pump speed by adjusting the available supplied power. This means identifying the pump suction state according to an intrinsic motor signal such as electrical current. However, this method did not prevent the suction condition, but it helped to prevent over-pumping [57].
Another approach has been developed where the pump is attached to one of ventricles of the heart, as a left or right assist device. This model consists of a ventricular assist device, cannula, controller, and the recipient’s cardiovascular system. For example, Yih-Choung et al. and Lim et al. [58] claimed that their esti- mator could identify cardiovascular system parameters such as atrial compliance, peripheral resistance, and inertial properties of the blood as inputs to the controller to facilitate an effective control strategy [57-59]. These variables could be obtained in principle by sensor technology, but because of the disadvantages of sensors, researchers have looked to find alternative methods. Existing models have also been used to investigate pump control strategies that can respond to hemodynamic changes. However, the need to transform sensor measurements, such as flow rate and pressure, into the desired cardiovascular variables is recommended. Hence, flow 42
and pressure are mathematically estimated through the use of non-invasive
measurements of motor current and pump speed.
2.3.4.2 Estimation of pump flow and pressure head in LVADs
The non-invasive estimation of LVAD flow rate and pressure has attracted the
attention of many researchers. Various pump control algorithms have been designed
by different research groups, operating under steady-state or transitory conditions,
who have shown that pump flow rate and differential pressure head can be
accurately estimated. Kitamura et al. showed that continuous measurements of pump
pressure and flow are not attempted in long-term clinical use of artificial hearts due
to disadvantages such as catheter blockage with a thrombus, or septicaemia in the
recipient [60]. The proposed method first estimates blood viscosity by linear
equation 2.3 (see below) and then solves a set of three equations for the pump
system to calculate flow rate and pressure head. The three equations are as follows:
where denotes pressure differential from the aorta through the pump to the left
atrium, is the output flow of the centrifugal pump, is the natural heart output,
is the total systemic resistance, and is the compliance of the aorta.
The equation governing the centrifugal pump is:
43
Where is the rotational speed of the impeller, is the total inertance of the blood within the inlet and outlet pump cannulae, and are viscosity-dependent parameters.
The rotor equation is given by:
Where is the DC-motor current, is the inertia of all the motor rotating parts, are viscosity-dependent parameters, is the kinetic friction coefficient of the rotor axis, and is the torque constant of the DC motor.
However, pulsatile flow estimation was covered by only a few studies. This may be
due to a number of factors, such as the use of pump characteristic curves which are sensitive to many physiological and mechanical parameters such as blood viscosity and pump impeller inertia. The design characteristics for each pump can contribute to the proposed estimation algorithms: this means that different unique equations can be obtained. Wakisaka et al. [61, 62] noted that it is important to determine the pump flow rate for a mechanical heart in clinical use after implantation in a recipient. Flow rate can be estimated by power consumption and the rotating speed of the motor, but because of fluctuations in blood viscosity, there is a need to understand the factors affecting this. He suggested a non-invasive method for correcting this estimation for centrifugal blood pumps (National Cardiovascular 44
Centre -NCVC-2™,Tokyo, Japan). Output power, pump speed and the haematocrit are measured, and their flow rate estimation method is shown in equation (2.4):
is the estimated flow rate (L/min), is the power consumption of the motor
(watts), is the rotating speed (rpm), and is the haematocrit (HCT). The values for parameters , and , were determined by least squares fits to data.
Tanaka et al. [63] proposed a new method for estimating instantaneous pump flow rate and pressure, based on motor voltage, current and rotational speed. They used two auto-regressive exogenous (ARX) models where the current and past values of power and speed were used as inputs to another ARX model. Their estimation met- hod has been validated on-line in acute animal experiments, with actual measured values of blood viscosity. Their method was successfully able to estimate pump flow and pressure when the blood viscosity changed.
Yoshizawa et al. [64] utilized power and pump speed to estimate pulsatile pump flow and pressure as the output of a two dynamic time series ARX models, as shown below in equations 2.5-2.6. This was performed by employing six fluid solutions, representing various levels of viscosity, as input parameters. After collecting the values of input and output, the coefficients were identified, the pulsatile flow and pump head pressure were estimated by employing two ARX models together. In the
45
first model, the coefficients were identified before pump implantation (off-line) using two inputs: pump speed and power. The second ARX model identified the steady state gain of the system, estimating blood viscosity on the basis of two inputs: power and rotational pump speed on-line (i.e. after implantation of the pump).
Furthermore, these two ARX models were combined to complete the estimation process with changing of blood viscosity each time and steady state gain recorded at these time changes. Furthermore, a mock loop system was used for the validation, with an error of 1.66 L/min which was affected by the input gain. The first ARX model was given by:
where is the electrical power, is rotational speed, and b1 – b6 are parameters to be identified by measured data. The second ARX model was given by:
where is the discrete time index satisfying , is the sampling period,
is the residual assumed to be white noise, is the order of the output, is the order of the input, are six kinds of exogenous input and and are coefficient parameters.
Furthermore, due to the insensitivity of LVADs to overflow and underflow, together with uncertainties in estimating pump flow, it is appropriate for pump control
46
strategies to prevent uncomfortable pumping conditions for the recipient. Malagutti et al. [65] investigated the role of the blood hematocrit (HCT) on a non-invasive iRBP flow estimation algorithm. Three input variables were used for flow estimation: the electrical power, pump speed and HCT. The resulting algorithm was then tested with a validation data set using different blood viscosities. Since iRBP flow depends upon the amount of blood in the LV, estimating blood flow and differential pressure is essential in designing an automatic, robust, and responsive control system that can effectively control the pump flow according to the body’s physiological needs and perturbations [66].
Most LVADs currently function at a fixed speed, which is controlled by the medicalspecialist. On the other hand, Bertram has found that detection of pump flow and differential pressure alone can be the basis of an adequate control strategy to assist the failing heart [66]. Wu et al. [67] based their algorithms on the control of aortic pressure rather than pump differential pressure. Other investigators have estimated pump pressure and flow in cardiac failure conditions, represented these as dynamic functions of the motor variables, and identifying the unknown coefficients mathematically. However, these coefficients depend on which specific pump and motor are used. For example, the fluid-dynamic design of the left ventricular assist device (VentrAssistTM) produces a particularly flat head-flow (HQ) curve [68]. Lim e t al. [69] proposed a dimensional analysis method for the estimation of implantable rotary pump flow and head pressure. This method utilizes theoretical principles of fluid mechanics, which provides a valuable understanding of the parameter relationships. Furthermore, their flow and pressure estimates were validated with two LVADs from VentrAssist™. Linear regression between the estimated and measured flow and head pressure values was plotted over a flow range 0.5 L/min to 47
8.0 L/min. Resultant slopes were 0.98 and 1.027 respectively, with respective errors of 5.79% and 1.51%.
Moreover, Ayre et al. [70] have used static, non-pulsatile and pulsatile mock loops with aqueous glycerol (AG) solutions and ovine blood to estimate average flow for a single viscosity. The static equation for the estimated flow was of the form:
(2.7) where Qest denotes estimated flow rate (L/min); denotes pump rotational speed
(rpm), and VI represents input electrical power (W). The values obtained for each of the six parameters (a, b, c, d, e, and f) were then plotted against blood viscosity, in order ascertain how to incorporate blood viscosity in a comprehensive relationship.
In this study, the authors proposed a non-invasive method to estimate iRBP flow in various hydrodynamic states using mock-loops and in-vivo experiments. Although these studies showed successful results when used to estimate the pump flow and pressure head, they did not focus on the stability of transient changes in pump flow.
However, Alomari et al. [71] proposed a dynamic model for pulsatile flow estimation of LVAD pressure and flow. The inputs of this model were acquired from the pump electrical power ( ), impeller rotational speed ( ) togther with HTC measured non-invasively. A viscosity range of 20% -50% was implemented in the mock loop using various AG solutions. Furthermore, a linear regression analysis between the measured and estimated pulsatile flow revealed a highly significant
48
correlation ( = 0.957) and mean absolute error of 0.902 L/min. In addition, further validations were undertaken using six sets of ex-vivo animal data. Their model provided accurate estimation of the transient response and the dynamics of pulsatile flow. The investigators utilized the steady-state solution given by equation 2.8:
where and are constants and the power coefficients and were found to have a linear relationship with the HCT. The dynamic version of the model defined an input to the dynamic system f(t) as follows:
f(t) =
where , is the sampling period and
The dynamic model for the left ventricular assist device was given by:
where is the output of the system which represents the estimated instantaneous value of the pulsatile flow is the input to the dynamic system model defined in Eq. 2.9, is the shift operator, whilst are polynomials defined as:
49
where is the model output order, and is the model input order satisfying condition .
In addition, AlOmari et al. [72] proposed a dynamic model for LVAD inlet pressure
estimation during the diastolic period. This model was used to design a controller to
regulate mean diastolic pump inlet pressure. Non-invasive variables such as pump
speed and power as well as a pulse width modulation signal (PWM) were used as
inputs to an ARX model to estimate inlet pressure during the diastolic period. The
model was validated using in-vivo animal data under a wide range of pump speeds
under various heart conditions.
2.3.4.3 Existing LVAD control strategies
Ventricular assist devices are used for long-term therapy for congestive heart failure
patients. Among these devices, iRBPs have been increasingly popular due to their
small size, and simple design. However, one of the disadvantages of rotary blood
pumps lies in their control strategy for maintaining outflow under different
physiological HF conditions. Most implanted LVADs are controlled manually by a
trained physician, using fixed procedures to adjust the outflow rate for patient
comfort. However, automatic and robust control systems have seen the steady
replacement of the role of clinicians to maintain physiological perfusion under
50
various conditions. This will hopefully contribute in helping LVAD patients live normal lives and continue their regular activities.
Efforts have been made to study and develop a variety of controllers for implantable ventricular assist devices. These can be classified according to controller input variables such invasive and non-invasive data. However, there are two designs of non-invasive controllers which have been used to control rotary blood pumps for maintaining physiological perfusion, which will be briefly explained. The invasive method of implanting more sensors is not recommended for long-term therapy, since additional sensors can result in inflammation and thrombus formation.
The first non-invasive control technique uses physiological variables as the inputs for the controller to determine the pulsatile and continuous pump flow rate. These variables include atrial pressure, blood oxygen saturation and lactic acid levels in the blood. Moreover, a number of physiological control strategies have been proposed for iRBPs, and many research groups have decided that maintaining atrial pressure within limits that balance blood volume in both sides of the heart is a practical method. Mclnnis et al. [73] proposed a physiological control system that incorporated LVAD and CVS models, which was controlled by a self-tuning PID- controller to control pulsatile blood flow [74]. This controller was designed to cope with different types of physiological demand to regulate left atrial pressure (LAP), and was validated by using mock-loop circulation system data. The controller
51
performed well when LV physiological conditions were simulated in the mock-loop.
The stroke volume (SV) of the assist device was adjusted by the control system so that the mean left atrial pressure remained at the desired level. Kitamura et al. proposed a new method to avoid using pressure sensors for measuring atrial pressure
[75]. The authors used a dynamic compensator technique for left atrial control for use with an artificial heart drive system. The goal here was to improve optimal pump output for the volume that the patient's circulatory system required.
Most recently, Karantonis et al. [76] proposed a control algorithm for an implant- able centrifugal rotary blood pump based on non-invasive indicators of the implant recipient’s activity level. The authors used the Activity Level Index (ALI) as input to the controller. ALI was based on two variables: tri-axial acceleration and heart rate
(HR), which together can be used to estimate metabolic energy expenditure. This non-invasive method makes it possible to control left ventricular stroke volume to compensate for changes in LV end-diastolic volume or its correlate preload.
However, Nakamura et al. [77] proposed mixed venous oxygen saturation
(SVO2) as an input variable for physiological control of a TAH. However, use of SV-
O2 sensors is not recommended because of the likelihood of thrombosis and inflam- mation in the patient.
The second non-invasive control technique to estimate pump flow and pressure is by using non-physiological input variables such as pump electrical current, voltage, and rotational speed. In this technique, flow and differential pressure used for the control algorithms are empirically estimated typically through experimental recordings of 52
pump flow and differential pressure. For instance, Giridharan et al. [78] proposed an effective proportional integral PI feedback controller to maintain a reference differential pressure between the LV and aorta. This model-based control system was able to regulate the pump speed within physiological limits, minimizing the difference between the reference pressure and the actual differential pressure.
Alomari et al. [79, 80] proposed a deadbeat controller to control pulsatile pump flow using a tested algorithm for steady and pulsatile reference pump flows. Pump speed, power and pulse width signals were used to estimate the pulsatile flow and inlet pressure. These values were used together with servo parameters as inputs to the control system, with pulsatile flow and pressure head as the output. In addition, both constant and sinusoidal reference pump flow inputs were used to test the controller algorithm under healthy and HF conditions. This controller showed that the simulated flow accurately tracked the reference input signal with an error of ±0.7
L/min [81].
Moreover, there are several varieties of non-invasive control methods which include suction limit control and preload control; all of which require demand-responsive control algorithms for implantable centrifugal pumps.
53
2.4 Existing LVAD Models and Computational Methods
2.4.1 Overview of Computational Techniques for LVAD-AV Modelling
Numerous computational methods have been used to simulate mechanical
interaction between blood and AV motion in the heart. This section will provide an
overview of fully-coupled fluid-structure interaction numerical techniques, whereby
fluid flow in the model affects mechanical defamation of solid structures and vice-
versa. This fully-coupled approach forms the basis of models developed in this
thesis.
2.4.1.1 Fluid-structure interaction
Fluid-Structure Interaction (FSI) was developed in the 1970’s to simulate the
interaction between a deformable solid in contact with fluid. FSI research initially
focused largely on aerospace applications, as reported by Farhat et al. [82]. It is now
used for a range of engineering applications dealing with the multi-physics
relationships between deformable solid structures with a surrounding and/or internal
fluid. To date, three main aspects of FSI methods have been developed. First, is the
technique used to discretise the governing equations for the solid and fluid domains,
which include the Finite Element Method (FEM) and Finite Volume Method (FVM).
The FEM will be discussed in more detail in the subsequent section of this chapter.
54
Second, is the coupling interface between the fluid and solid. For cardiovascular models involving structural motion change (e.g. AV deformation) due to fluid flow
(e.g. blood flow), the FSI method is very practical and accurate in simulating native heart physiological behaviour in terms of wall contraction and valvular conditions during the cardiac cycle. The computational models developed in this thesis are governed by the complex interaction between the heart wall, valve leaflets and blood flow. The coupling method used to link cardiac geometry with blood on the solid- fluid interface is a key issue. Existing coupling methods are categorized into two categories, partitioned and monolithic. The one-way method, used by most FSI approaches, is a partitioned approach [83]. The monolithic method represents a strong-coupling approach, a two-way mathematical or computational process which relates the solution method with the movement of the solid structure and thereby affects the blood flow which, in turn, affects the motion of the structure. The partitioned method, is weakly-coupled, separating the fluid and solid structure and then solving their governing equations with two different solvers until both solutions converge.
The next section will provide a brief overview of the finite element method, which was used to numerically implement the FSI models of this thesis.
55
2.4.2 Finite Element Method
Finite element methods are employed extensively in the analysis of solid structures
and fluids, and indeed are useful in almost every field of biomedical engineering
design. FEMs are the numerical tools of choice for solving partial differential
equations (PDEs) and are widely used in biomedical engineering applications, with
this use expected to increase significantly in the coming years. Complex
physiological systems require appropriate methods for structural analyses, with
numerical techniques for structural analyses such as FEM and boundary element
methods (BEMs), are promising approaches used from the 1950s, with the help
of advanced computer systems [84].
Accurate prediction of fluid dynamics and solid mechanical response is essential in
the simulation of native heart valve dymanics, particularly, the opening and closing
behaviour of the aortic valve during LVAD support. Finite Element Analysis, or
FEA, involves using mesh generation for dividing a complex problem into small
elements with the use of computational software programs coded with FEM
algorithms. Domain meshing is an important step in the prediction of native heart
responses and potential abnormalities in heart valves (i.e. thromboembolic
complications and AV stenosis) to LVAD support, with underlying physics such
as the Navier-Stokes fluid equations given as PDE or integral equations solved over
each element.
In addition, simulation methods are highly useful and efficient, since in-vivo
experimental approaches often require the creation of expensive, time-intensive 56
prototypes and evaluations. Therefore, numerical methods such as FEM are often preferred for investigating complex dynamic systems, including the interaction of the valves with the blood to understand their function in the human heart during
LVAD support.
FSI when combined with FEM, aims to simulate the motion of solid structures (i.e.
AV leaflets) due to a surrounding fluid. In contrast, in purely structural FEM, there is no fluid-solid coupling; the loads must be assigned before the solution can be found. Using FEM, the irregular geometries of human organs or tissues are replaced by discrete models developed by subdividing into a number of finite elements. These discretised, as an assemblage (mesh) models, are composed of appropriately shaped elements represented by a chain of interconnected points known as nodes. Elements can be rectilinear or curvilinear, forming a grid. Implicit within each element is its force/displacement function which, in terms of parameters to be determined, defines how the force/displacements (i.e. element stiffness properties) are interpolated from the nodes over each element. This may be either an external force applied as a load to the structure, or an internal force that transfers loads between elements. The equations of equilibrium are solved for each element, and variables in time can be modelled using either implicit or explicit methods by dividing into a number of time steps. When applied to the analysis of a continuum fields over a domain, either solid or fluid, the spatial discretisation forms a matrix assembly consisting of a number of degrees of freedom. Using the implicit method, acceleration is assumed to vary linearly within each time step, which gives a more accurate representation. In the
57
explicit technique, the equations of motion are solved at every time step by assuming
that acceleration is constant over the step.
A major advantage of FEM is that relatively moderate finite degrees of freedom can
model a range of complex forms (i.e. one or two dimensional shells or solid
elements), and can be handled with ease, and numerically solved with system
algebraic equations. The essential characteristic of FEM is that the solution of the
discrete system is assumed, a priori, to have a prescribed form. Also, the solution is
strongly linked to the geometric representation of the domain, which is an integral
(or weak) form of the PDE. Generally, this integral (weak) equation is developed
using a weighted residual method which can include differential type boundary
conditions. A final characteristic of FEM is the modular way in which the
discretization is achieved, which is assembled from an element-element level to
define a global matrix system of equations [85].
The next section will provide a brief overview of existing computational studies of
ventricular support devices and their interaction with the cardiovascular system. This
thesis aims to extend the scope of such models to incorporate AV motion.
2.4.3 Heart-Pump Interaction Models
Implanted LVADs are difficult to monitor using standard medical image modalities
such as MRI and echocardiography due to limitations with device positioning as
58
well as the metallic parts of the pump and impeller. Consequently, numerical simulations using CFD and FSI schemes solved with the Arbitrary Lagrangian–
Eulerian Finite Element Method have been important tools for cardiac function research, providing a crucial visualisation of cardiac function, blood haemodynamics and abnormalities in the circulatory system. Since this approach is based entirely on non-invasive tests using computer software, issues of reduced experimental reliability and relatively higher costs associated with incorporating in vivo tests are avoided.
Non-invasive experiments such as computational simulations, have emerged as key tools for directly analysing the complex mechanics and dynamics of the heart. Such numerical simulations have been used ny many investigators for evaluating heart function, simulating the dynamic changes in the cardiovascular system (CVS) under healthy, diseased, and VAD-supported conditions. Such numerical models could be used to simulate CVS behaviour in the presence of an LVAD, providing additional insights into the dynamics of assisted circulation. This could provide insights into the performance characteristics of the blood pump and its interaction with the pulsating heart. Non-invasive computational methods of heart-pump interaction can be categorized into lumped parameter and finite element models.
59
3 Chapter 3-Review of Modelling Approaches in Cardiac Dynamics
3.1 Background
Understanding the fundamental response mechanisms of fluid/solid coupling in the
heart is crucial for characterizing normal heart function and its behaviour in disease.
Although flow in the human heart, particularly the left ventricle (LV), has been
investigated by invasive and non-invasive imaging [86, 87], there is still only a
limited understanding of blood and heart tissue interaction, particularly if the LV is
assisted with a pump device. The latter scenario is the basis for the modeling work of
this thesis.
Lumped parameter models have increased our knowledge and understanding of
cardiac function and the interaction between the LV and its valves [88]. FSI is
commonly used as an important modelling approach to acquire an understanding of
cardiac hemodynamic and heart valve behavior [89-91].
In recent years, cardiac CFD simulations, including studies of blood flow patterns,
have played an important role in investigating heart wall motion and valve function.
More detailed understanding of the coupled relationship between blood flow patterns 60
and the myocardial valves can be further enhanced by the combined use of CFD and patient specific geometries. It represents a reliable tool that can be used to improve our understanding of heart changes during the cardiac cycle such as relaxation and contraction, as well as the pathophysiology of heart valves, which could be very useful in the investigation of AV status during LVAD support. AV dynamics has been largely ignored in previous modelling studies of the LVAD-supported heart.
Although LV-pump interaction may be partially understood through in-vivo animal studies, such studies are unconvincing at present due to a shortage of animal models of heart failure [92]. The primary advantage of computational models is that they play a significant role in simulating heart failure experiments under similar conditions. Therefore, such models offer an excellent platform for the development and evaluation of cardiac physiological dynamics.
On the other hand, there are three types of in-vitro based methods to simulate blood flow within the ventricles to reconstruct the mechanical behaviour of the LV and aortic valve: the geometry-prescribed CFD method, the fictitious FSI method, and the realistic FSI method. These methods offer full flow features in the ventricle, such as velocity profiles, pressure, shear rate and recirculation. They have been commonly used in the simulation of heart valve function during cardiac cycle [93].
In this thesis, the realistic fluid-structure-coupled simulation of the AV and left ventricle during ejection phase under LVAD support is utilized.
61
Realistic FSI requires at least three aspects to accurately simulate heart dynamics, as reported by Cheng et al. [94]. First, the structure solver must be able to define the nonlinear, anisotropic and inhomogeneous tissue characteristics. Next, the CFD method must be able to solve the large deformations at the boundaries of the fluid domains, updating the meshes accordingly. Last and most importantly, the coupling algorithm must be able to couple the fluid (i.e. blood) and structural elements (i.e. heart valves and inner walls) correctly and ensure convergence. In this thesis, this approach is applied to specifically investigate LV fluid/solid coupling, allowing quantitative examination of blood flow through the LV, pressure distributions and
AV movement.
The survey of literature presented in this chapter, regarding cardiac dynamics modelling, will focus on two aspects. The first is a review of CFD studies which have been undertaken using realistic simulations of the LV, including simulating AV dynamics. The second will examine studies that have simulated AV function under
LVAD support. However, none of these models include a two dimensional (2D) model of AV state during LVAD support under pulsatile flow conditions, as presented in this thesis.
62
3.2 Existing cardiac dynamics models
Numerical simulations of the mechanics of the heart may be divided into three
general categories, solid-only, fluid-only, and fluid-structure interaction. Many
studies have focused on solid-only simulations that compute the deformations of the
heart walls by specifying inner boundary conditions for the pressure [95] [. Fluid
only simulations simulate the flow of blood through the heart by assigning prescribed
wall motions in patient-specific geometries, according to data collected from
magnetic resonances edition imaging (MRI) techniques [96].
ALE methods employ a grid that is modified to, and deforms with, the moving
boundary. It has many advantages for finding solutions to a wide range of time-
dependent complex fluid dynamics problems such as three dimensional (3D)
simulations and studies of blood flow in arteries [44]. In addition, the method uses a
finite difference mesh with vertices that may be moved with the fluid (Lagrangian
frame), be held fixed (Eulerian frame), or be moved in any other prescribed manner.
The theoretical framework for mixed Lagrangian-Eulerian finite element
descriptions has been established by Hughes et al. [97] in the context of
incompressible and viscous flows.
63
A large number of investigations have been carried out on the haemodynamics of the
left ventricle using CFD techniques. This computational method has been widely
used since the early 1970s: for example, Bellhouse et al. [98] provided a LV model
incorporating the mitral and aortic valve to investigate the effect of LV vortices in
mitral valve open and closed states. Reul et al. [99] investigated the forces causing
mitral valve closure under an adverse pressure gradient, incorporating valve leaflet
motion. Taylor et al. [100, 101] simulated haemodynamics within the left ventricle,
including detailed flow and pressure gradient behaviour. Their computation used a
3D realistic LV anatomy. Jones et al. [102] extended this work to more accurately
simulate blood flow within the LV. Vierendeels et al. [103] developed a 2D axi-
symmetrical model of detailed LV flow during filling, examining the flow patterns
produced during the diastolic phase. These studies have provided important results
on the role of the LV wall stiffness on blood flow during the filling cycle.
In addition, pioneering 2D and 3D simulations in cardiac dynamics include the studies of Peskin and McQueen [104-109]. These investigators incorporated a left chamber into their heart models using the immersed boundary method. The solid domain was not explicitly characterized within the fluid field, however it was represented by an additional force field that the solid applies on the fluid where the two fields overlap.
As a result, the first model of left ventricular valve function using the immersed boundary method was due to Peskin [110], who detailed the fluid-ventricular wall interaction scheme [105].
64
Vesier et al. [111] have developed a 3D ventricular thin-walled model, which gives a realistic behaviour of LV function when compared with clinical data. In addition,
Lemmon et al. [112] have used the same realistic approach to cardiac modelling with the immersed boundary method in a 3D heart model to simulate the blood-tissue inter- action. This technique offered a significant tool in various studies to assess cardiovas- cular function and investigate the phenomena of vortex and swirling motions within the LV interacting with heart valves [113-115], inflow velocity and pressure, particularly intraventricular flow dynamics on mitral [116] and aortic valve [117]
flow. Long et al. [118] determined that CFD simulations of left ventricular flow are highly sensitive to the boundary conditions imposed. In addition, Vierendeels et al.
[103] proposed a 2D axisymmetric ventricular model based on the ALE method, to examine the formation of pressure and vortex patterns during the filling phase.
These simulations did not take FSI into account. FSI needs to be carried out using different, albeit tightly coupled, sets of equations for the fluid and solid domains, using either a monolithic or a partitioned coupling method [119, 120]. The differences between these is that the monolithic method solves two sets of governing equations (i.e. solid and fluid) at the same time and is computationally challenging.
However, the partitioned scheme computes the solid and fluid fields individually, coupling these together through iterative updating of boundary conditions at the interfaces. Consequently, it can minimize the computational process time by taking advantage of sophisticated codes that have been developed to solve pure fluid or structural problems. Borazjani et al. [121] have identified the relationship between the background grid and the moving bodies. They conclude that the stability
65
of FSI code depends both on the properties of the structure (mass and geometry) and on the sign of the local progressive force imposed by the local flow on the structure.
Hence the FSI method, which is more accurate and realistic, directly couples both the solid wall motion of the heart and the blood flow. For example, Chahboune et al. [122] proposed an FEM based FSI scheme for 2D simulation of blood flow inside the LV over a complete cardiac cycle. Patterson et al.[123] and Carmody et al. [11] have developed a significant 3D FSI model of the mitral valve, known as the Sheffield bicuspid valve model. Their studies have included the simulation of blood pulsatile flow for one cardiac cycle along with LV wall motion using distinct methods, using the outlet flow pattern as the input for AV and aortic root flow profile. Their LV mod- el was geometrically built from MRI data, and results corresponded well with physiol- ogical AV opening times and flow velocities. Vierendeels et al. [124] coupled the fluid and solid equations to simulate haemodynamics for a 2D axisymmetric LV. Their
simulated blood flow field was achieved using the ALE method combining the full
Navier-Stokes equations on a moving mesh and computed from the movement of boundaries developed by Riemslagh et al. [125].
In addition, De Hart et al. [18, 126-128] published an important FSI study of the aortic valve, which incorporated a 2D AV model. Their method was focused on solving for the fluid system variables first (i.e. pressure and stress), while the structural variables
(i.e. displacements) were kept constant. They investigated the application of strong coupling to the problem of a flexible aortic valve during systole. Their FEM m-
66
odel utilized the fictitious domain technique to couple the fluid and structural domains.
It was extended to a 3D model in later studies [18, 128].
The model of Hunter et al. [129] goes a long way towards achieving the whole-organ model objective of integrating cardiac anatomy, electrical activation, mechanics, metabolism and fluid mechanics together, although their model does not incorporate
AV leaflet motion. Vierendeels et al. [124] proposed a 2D axisymmetric FSI model to simulate passive filling of the ventricle, providing pressure and vortex pattern results.
Loon et al. [130] link fictitious domain (FD) with adaptive meshing for flexible valve leaflets. Stijnen et al. [131] use FD to predict the dynamic behaviour of a 2D moving rigid heart valve .
Watanabe et al. developed a three dimensional, FEM-based model which incorporated
blood flow inside the LV using FSI, examining the contraction and relaxation of the
LV during the cardiac cycle in normal activation as well as arrhythmia [132, 133].
Cheng et al. [94] developed an FSI method to analyse the velocity, vortex and pressure distributions under LV contraction and relaxation, regardless of the interaction with the mitral and aortic valves. Kittian et al. [120] modelled a LV FSI model, including both passive inflation and active contraction, based on patient-specific MRI datasets to evaluate the quality of pressure distribution results, which showed an overall agreement with experimental data.
67
However, the above models did not include simulation of the closing state of
the aortic valve during LVAD support, and ignored its interaction with the blood and
solid wall structures. Specifically, how does the AV behave under LVAD support?
Most of these studies support only the design of prosthetic valves and its influence on
surgical techniques required to repair mitral and aortic valves. In addition, none of
these studies provide full FSI models of the AV during LVAD support.
3.3 Existing modelling studies of AV state under LVAD support
Aortic valve state forms an important consideration in regards to thrombus formation
in assisted hearts due to its influence on intraventricular flow distribution in the LV.
Simulating the interaction between the blood and AV is challenging due to the
complex dynamics of the AV, the potential contact between AV leaflets, basic flow
instability, and intense velocity and pressure gradients local to the AV. However,
many CVS-LVAD interaction studies have been undertaken to investigate the impact
of rotary pumps on heart valve function. Lim et al. [86, 104] have developed a
lumped-parameter model of the CVS based on in-vivo experimental data recorded in
healthy animals combined with a dynamical model of a left ventricular device. This
model was used to evaluate the impact of LVADS on heart dynamics, including heart
valves.
68
Another approach to in-vitro experiments is to use a mock circulation loop system
(MCLs): this has been employed by various research groups to mimic LVAD-heart valve interactions [134]. A study by Shi et al. [135] presented a numerical model of the CVS, including the AV, under the pumping action of a pulsatile VAD connected in-series with the native heart. Their model was used to predict the change in aortic pressure (Aop) and coronary flow under different physiological conditions with
LVAD support. In addition, Endo et al. [136] have estimated the AV state during pump support using a closed-loop mock circulation with pulsatile pump simulating the natural LV.
In terms of numerical simulations, the Arbitrary Lagrangian Eulerian method
(ALE) is able to provide a suitable solution of the Navier–Stokes equations for fluid
flow with structural interactions characteristic of AV simulations. However, due to the large leaflet deformations, as reported by Horsten [137], it is important to adapt the fluid-domain mesh in such a way that proper mesh quality is maintained to model the
AV motion. We have used the ALE finite element method in our 2D AV models of this thesis to simulate leaflet motion with blood flow.
Accurate simulation of fluid/solid coupling mechanisms are crucial for characterizing normal heart valve function and its behaviour during LVAD support. To investigate these mechanisms, Nordsletten et al. [138] implemented an FSI LV model by integrating multiple physiological data to characterize the haemodynamics of the LV, including passive diastolic and active systolic phases. Su et al. [84] modelled the first
69
patient-specific left ventricular flow in two-dimensions, taking into account both the mitral and aortic valves.
Understanding how the AV behaves under LVAD support may allow the detection of its open/closed state from LVAD pressure/flow sensor transducers alone, paving the way for more sophisticated pump control algorithms which take into account such AV state. A particular problem of interest here is the effect of outlet pump outflow on AV closure of the human heart assisted by an implantable blood pump, and how it could be estimated from non-invasive pump variables. The aim of the computational models of this thesis is to simulate the interaction between the LV and aortic valve under
LVAD support. It is clear that this is a challenging task that needs to be addressed one stage at a time.
70
Part II – Methods
71
4 Chapter 4 -Development of Computational 2D LV-pump Models
4.1 Introduction
Many computational studies have been undertaken to simulate the behaviour of the
aortic valve, ignoring its interaction with the blood, during LV support [89-91].
However, several studies have also been published incorporating full-FSI of the AV
leaflets in contact with blood [18, 131, 139-141]. This chapter presents the
methodology we have developed in this thesis for a 2D LV-pump-AV model using a
simplified LV chamber with AV leaflets and LVAD blood flow. Although recent
imaging-based works can produce 3D patient-specific geometry over time, this
thesis have focused on more simpler 2D models to better understand the LVAD
factors affecting AV state during the cardiac cycle. These 2D simplified models
represent important pioneering steps on the way to more refined models, including
those based on patient-specific data.
Initially, our simplified LV dimensions were adapted from the study of McQueen et
al. [109] however we later modified the LV geometry to a more-realistic shape with
appropriate measurements from the literature. COMSOL Multiphysics (COMSOL
AB, Sweden, Version 4.3a) was used to develop the geometry models and the FEM 72
meshes of the fluid and solid domains. The more-realistic model represents a 2D section through the left side of the heart and is based on a study by Peskin [142]. Our purpose here is to modify the model’s dimensions and parameters so it mimics a human LV and to validate simulation results with previous studies. The more- realistic model describes blood flow in the LV and the movement of heart walls during a cardiac cycle. We also generalised the boundary representation so that it included not only the LV cavity and AV leaflets but also the LV walls.
The computational meshing process was unique to the individual geometries, involving mesh parameters such as element type, size and resolution of elements in high-curvature regions. The modelling process included specifying the model geometry, assigning the material properties, and applying appropriate boundary conditions. Our models aim to illustrate how fluid flow (i.e. blood) can deform the
LV and AV leaflet structures, solving for the flow in a continuously deforming geometry using the Arbitrary Lagrangian-Eulerian technique.
Fluid-structure interaction (FSI) deals with the multi-physics relations between deformable solid structures and a surrounding and/or internal fluid. FSI deals with structural motion change (i.e. LV wall and AV leaflet deformation) due to fluid (i.e. blood) flow, and is ideally suited to simulating AV behaviour in terms of opening and closing times during LVAD support. This behaviour is governed by the complex interaction between the valve leaflets, blood flow and blood pressure, as well as 73
LVAD pumping action. In FSI simulations, the coupling method used to solve for
deformations and flow at the boundary interface is a key issue. FSI coupling
methods are categorized into a three approaches: weak, strong and constrained met-
hods. COMSOL FEM software by default is set to fully-coupled (i.e. strong).
FSI simulations were performed for two LV geometrical groups. For each group, A-
V state was detected and analysed for various modelling scenarios. First we simulat-
ed the LV using a simplified rectangular chamber, whilst for the second group, a rea-
listic LV geometry was implemented based on previous literature. These simulation
scenarios were investigated under various LV dynamics, including the presence of
systemic circulation and LV contraction.
In this chapter, we define each group’s dimensional and numerical settings,
including the LVAD pump model. Adapted geometric methods have been used for
various purposes:
i. to detect AV status during LVAD support.
ii. to examine the resulting data for pump current, average aortic outflow
waveforms and distance between AV leaflets under various pump speed set
points and physiological conditions. iii. to estimate average pump differential pressure and average pump flow rate
from average pump input power.
74
The last part of this chapter describes the simulation protocols used to investigate
the LV-pump interaction model over a wide range of physiological operating condit-
ions, including variations in:
i. systemic vascular resistance (SVR); and
ii. total blood volume Vtotal by varying the inlet-flow rate to LV model.
The above variations are common in real life scenarios, and could lead to adverse
patient states with intolerable risk.
4.2 Geometrical models of LV-LVAD interaction
4.2.1 Simplified LV-pump interaction model 4.2.1.1 Computational approach and geometric definition:
The geometry of the simplified 2D LV-pump model was chosen to reflect that of
simple LV dimensions, with the domain having the same dimensions as a slice
through the measurement section. The ALE-based finite element method was used to
implement a fluid-structure interaction solver, which allows for deformation of
flexible structures such as the valve leaflets within the LV chamber. Our objective
was to analyse the motion of the AV in a highly-simplified LV model, without
taking into account the effects of heart wall contraction.
75
Figure 4 .1: Simplified 2D representation of the LV, aortic valve and pump cannula.
The model geometry consisted of a horizontal flow channel of diameter 2.0 cm and length 3.5 cm representing the LV, an appended 2.5 cm length representing the aorta
(Ao), and narrow curved structures representing the AV leaflets, as shown in Figure
4 .1. The AV consisted of two flexible leaflets of length 1.25 cm, idealized as rounded arcs with rounded tips. Although a native aortic valve has three leaflets, only two leaflets of equal length were modelled here, with assumed 0.1 cm thickness. To maintain a continuous flow domain, a 0.2 mm gap between the leaflets was preserved at the fully closed position (Figure 4 .2 (b)). A thin-walled cannula of width 0.8 cm and length 0.8 cm, representing the LV-AD inlet, was inserted into the simplified ventricular chamber wall. The model also included blood outflow from the LV: aortic outflow was modelled using four outlet boundaries defining a square- shaped ‘sink’, as illustrated in Figure 4 .1. Inflow to the LV was produced from four 76
inlet boundaries in the form of a square ‘source’, placed at the left end of the channel
(Fig. 4.1).
4.2.1.2 Mesh Generation
For most applied problems in fluid dynamics, it is essential to solve the governing
equations numerically with the assistance of computers. The formulation of the
equations into a form for numerical solution generally requires that the flow and
structure domains be discretised using a set of points or nodes which can be
connected to produce elements.
To obtain accurate simulation results, and using special meshing tools provided by
COMSOL, we combined two types of meshing methods and resolutions in the
model, as shown in Figure 4 .2,. These were a “fine” resolution for the fluid domain,
with maximum and minimum element sizes of 0.98 mm and 0.028 mm respectively,
and a “free triangular” mesh resolution for the remaining domains.
Solution accuracy with the finite element method is known to be influenced not only
by the element size, but also by the element quality – a measure indicating how close
the triangular or quadrilateral elements are to regular polyhedrons. A number of
measures are in use for the quality of an element, the most obvious being the
smallest and largest angles associated with the element [143]. In COMSOL, the
77
Figure 4 .2: (a) a fluid and structural element meshes in the simplified LV-pump model. The outer and inner AV segment lengths are 13 and 15.3 mm, respectively. (b) zoomed view of the AV highlighting the leaflet thickness, AV mesh and chamber element details.
mean element quality ( ) of triangular/ quadrilateral meshes was calculated by
dividing the area of each element by the sum of squares of its sides. The element
quality (q) is then scaled to range from zero to one such that an equilateral triangular
or square element has a quality of one. Therefore, for a triangular element we have:
and for a quadrilateral element:
78
where A is the area, and h1, h2, h3, h4 are the sidelengths of the element. The mean
element quality of the triangular and quadrilateral elements used ( =0.95 and 0.45
respectively), were significantly greater than the value known to affect the
accuracy of a solution (see Figure 4 .16), as determined by COMSOL to be =0.3
for triangular and quadrilateral meshes [85].
During movement of the leaflet structures, the nearby triangular elements will
deform according to COMSOL's in-built moving mesh interface, so that the mesh
quality remains as high as possible. If however the triangles do degenerate, there is
an option in the interface to perform a remesh if the overall mesh quality falls below
a threshold value. This option, however, was not required in the simulations of this
thesis, but could be implemented if larger deformations were imposed onto the
model.
4.2.1.3 Material properties
The viscosity of blood and elastic properties of the valve leaflets were obtained from
the literature to represent healthy human blood and valve properties [144, 145]. The
properties of blood used in this simulation are shown in Table 4 -1.
79
Table 4 -1: Material properties used for the simplified LV model
Parameter Value
Blood density ( ) 1.06 × 103
Blood viscosity (Pa.s) 2.70 × 10-3
Leaflet density ( ) 1.06 × 103
Aortic leaflet Young’s 1.0 × 106 modulus (Pa)
Leaflet Poisson’s ratio 0.49
4.2.1.4 Model equations
A two-dimensional FSI model of aortic valve dynamics was implemented using the
COMSOL Multiphysics finite element numerical software platform (COMSOL AB,
Sweden, Version 4.3a). The software utilized the moving grid method, whereby the
fluid mesh was adjusted to move with the moving solid-fluid interface boundary
throughout the numerical calculation. This technique is known as the Arbitrary
Lagrangian-Eulerian (ALE) formulation, first proposed by Donea et al. [146],
coupling the Lagrangian and Eulerian descriptions of the Navier-Stokes equations on
the moving grid.
Blood pulsatile flow was characterized as laminar, Newtonian, viscous and
incompressible. The fluid was described by the Navier-Stokes equations (4.1) and
the mass continuity equation for incompressible flow (4.2). 80
(4.1)
(4.2)
where is the fluid density, is the velocity of the fluid in the fixed Lagrangian coordinate system, is the velocity of the fluid in the moving Eulerian coordinate system, is the pressure, is the fluid viscosity, and I is the unit tensor.
The fluid flows into the LV chamber from the left source (inlet) boundaries (see Fig.
4.1). At this inlet, the flow is assumed to have fully developed a laminar profile, changing with time as described in the next section. The valves are modeled as a linear elastic material, formulated as an isotropic Hookean elastic solid expressed using Einsteinian indicial notation as:
(4.3)
where is the Young’s modulus, is Poisson’s ratio, are the strain and stress tensors respectively, and represent the Kronecker-delta tensor such that:
Values of E and for the AV leaflets are given in Table 4 -1
81
4.2.1.5 Boundary Conditions
The fluid flow boundary conditions of the model included the following: laminar
inflow conditions at the inlet boundaries were applied using a sinusoidal velocity
profile with a period of 1 s and amplitude 0.12 m/s, as shown in Figure 4.3, a 0
mmHg pressure outlet was specified at the sink outlet boundaries, a 0.014 m/s
velocity was applied at the outflow of the pump cannula, the fixed walls of the
model were set to be no-slip boundaries, and the valve leaflets were assigned a
'moving-wall' boundary condition whereby the velocity of fluid at these leaflet
boundaries was set equal to the velocity of the moving wall. Finally, the stress on the
valve leaflets was set to equal to the fluid stress (pressure plus viscous stress), with
the leaflet root boundaries set to a fixed displacement of zero.
Figure 4 .3: Sinusoidal velocity profile multiplied by an instantaneous step starting at 0.2 s applied at the inlet to simulate inflow.
4.2.1.6 Computational settings
The FSI simulations were performed on a computational workstation using a 64-bit 82
windows platform with 3.20 GHz processor and an Intel Core i7-3930K processor,
with an applicable memory allocation of 32 GB. The fully-meshed model exhibited
approximately 15,000 degrees of freedom.
4.2.1.7 Numerical Settings
The simplified LV pump model was implemented over two mesh geometries using
one setting for the integration time stepping algorithm: namely, absolute tolerance
0.001, relative tolerance 0.01, maximum integration time step 0.01s, maximum order
of the Backward Differential Formula (BDF) integration scheme at 2, saved sample
rate of the output at 100 Hz.
4.2.2 Simplified LV-pump interaction model with systemic circulation
4.2.2.1 Computational approach and geomtry definition
LVAD cannula placement influences the dynamics of flow within the ventricle,
however, placing the cannula near the centre of the LV leads to high velocities in
order to ensure ventricular washout as reported by Laumen et al. [18]. In our model,
the LVAD cannula was placed in the center of the ventricle between the inlet
boundaries (source) and the AV as shown in Figure 4 .4.
83
Figure 4 .4: (a) Fluid and structural element meshes of simplified LV with pump and CVS model. The length and diameter of the pump cannula are 0.6 and 0.6 mm, respectively. (b) Zoomed view of the AV highlighting the leaflet thickness, AV and chamber mesh element details.
The simplified LV-pump CVS interaction model geometry consisted of a rectangular
flow channel of diameter 2.5 cm and length 5.0 cm, representing the ventricle, along
with narrow curved structures representing the AV leaflets (see Figure 4 .4). The
latter were two flexible leaflets of length 1.65 cm and thickness 0.1 cm. A thin-
walled cannula of width 0.6 cm and length 0.6 cm was also placed into the
ventricular chamber wall. The model included blood outflow from the LV, which
was sourced from inlet boundaries placed at the left end of the chamber (labelled
“source” in Figs 4.4 and 4.5), whilst aortic outflow was modelled as a fixed pressure
boundary condition. Outflow from the aorta "sink" boundaries was calculated using
a Windkessel model of the circulation, connecting the aortic "sink" boundaries to the
atrial "source" boundaries, as shown in Figure 4 .5. In this version of the simplified
model, heart wall contraction was not considered.
84
Figure 4 .5: Windkessel model of the circulation where is the left ventricular outlet pressure, is the arterial systemic pressure, is atrial inflow to the LV, is the pump flow rate (L/min), is the blood flow ejected from the LV, is the characteristic aortic impedance, is the peripheral resistance and is the arterial systemic compliance.
In addition, blood pulsatile flow was characterized as laminar, Newtonian, viscous
and incompressible, and described by the incompressible Navier-Stokes (4.1) and
mass continuity (4.2) equations. The fluid flows into the model from the left source
(inlet) boundaries. At the entrance, the flow is assumed to have fully developed a
laminar characteristic with flow changing with time as described in the subsequent
fluid boundary condition section (4.2.2.5).
4.2.2.2 Mesh Generation
Three types of meshing algorithms and resolutions were used in this version of the
model: an “extremely fine” mesh setting was used for the fluid domain with
maximum and minimum element sizes of 0.80 mm and 0.0016 mm respectively.
Furthermore, due to the importance of simulating an accurate distance between the
85
leaflets to identify AV open and closed states, a small element size around the
leaflets tips (see Table 4 -2) was required. Therefore, AV boundaries adjacent to the
blood were discretised with 8 boundary layers and with 1 thickness adjustment
factor, producing thin quadrilateral elements parallel to the boundary. Finally, a
“free triangular” mesh was used for the remaining domains (Figure 4 .4). The fluid
and solid domains were discretised with triangular elements, as summarised in Table
4 -2. The total number of triangular and quadrilateral mesh mesh elements for this
model were 11696 and 2816 respectively.
Table 4 -2: Mesh properties of simplified CVS LV-pump model.
Regional of interest Simplified models with system circulation
valve leaflet edges Number of edge elements: 176
Mean element size: 0.58 (mm)
valve leaflet tips Number of edge elements: 32
Mean element size: 0.09 (mm)
Fluid domain Number of elements: 12236
Triangular: 10828
Quadrilateral: 1408
Mean element size: 0.75(mm)
Valve leaflet domains Number of elements:2110
Triangular:702
Quadrilateral:1408
Mean element size: 0.61(mm)
86
4.2.2.3 Material properties
In this version of the model, the viscosity of blood and elastic properties of the valve
leaflets were similar to that of the previous model (see Table 4 -1). However, for
improved accuracy the AV leaflet Young’s modulus was increased to 30 × 107 Pa.
4.2.2.4 Model equations 4.2.2.4.1 Windkessel model
To simulate the systemic circulation, a windkessel model was employed [147],
characterized by the following equations (see Figure 4 .5):
(4.4)
(4.5)
+ (4.6)
where P is the aortic pressure, Pout is the left ventricular outlet pressure, PS is the
arterial systemic pressure, is the pump flow rate, is the blood flow ejected
from the aorta, Rout is the characteristic aortic root impedance, RS is the total periph-
eral resistance and CS is the arterial systemic compliance. The following values we-
re employed for these Windkessel parameters: =0.006 mmHg.s/ , =1.0 m
87
mHg.s/ and =2.75 /mmHg These parameter values are adapted from
Danielsen [148].
4.2.2.5 Fluid boundary conditions
Boundary conditions applied to the model included the following: laminar inflow conditions were applied at the inlet source and outlet sink boundaries. For the inlet boundaries, a sinusoidal velocity profile with offset, having a period of 1 sec and maximum and minimum magnitudes of 1 m/s and -0.5 m/s respectively, was applied as shown in Figure 4 .6. For the outlet boundaries, we have used the Windkessel model to determine the pressure (Pout). For the LVAD, a fixed 2 m/s average velocity was applied over the outflow boundary of the cannula. Finally, the LV walls of the model were set to be no-slip boundaries.
Figure 4 .6: Sinusoidal velocity profile applied to the inlet boundaries to simulate diastolic inflow. The inflow velocity oscillated in a sinusoidal pattern with period 1 s and a magnitude 0.75ms−1 about a mean level of 0.25 ms-1.
88
4.2.3 Simplified LV-pump model with systemic circulation and heart contractility
4.2.3.1 Computational approach and Geometry definition
In this version of the model, the geometry similarly consisted of a horizontal flow
chamber of diameter 2.5 cm and length 5 cm representing the ventricle, and narrow
curved structures representing the AV leaflets, as shown in Figure 4 .7. Key model
parameters shown in Table 4 -3 were: blood density, blood viscosity, leaflet density,
aortic leaflet Young’s modulus and leaflet Poisson’s ratio. The AV consisted of two
flexible leaflets of length 1.25 cm and thickness 1 mm. A thin-walled cannula of
width 0.6 cm and length 0.6 cm, representing the LVAD inlet, was placed in the
ventricular chamber wall.
The model also included blood inflow into the LV on boundaries placed at the left
end of the channel (labelled as "source" in Figure 4 .7), whilst aortic outflow was
modelled as a varying pressure boundary condition (labelled "sink" in Figure 4 .7).
Outward flow at the atrial "sink" boundaries was simulated using a
Windkessel model of the circulation as previously defined, connecting aortic
pressure and flow. These equations governing the Windkessel circuit were
implemented as global differential equations in the COMSOL solver. Heart wall
contraction on the upper boundary of the ventricular chamber was also incorporated,
labelled as “moving-wall” in Figure 4.7.
89
Figure 4 .7: (a) Fluid and structural domain meshes for the simplified LV-LVAD interaction model with heart wall contraction. The length and diameter of the pump cannula were 0.6 and 0.6 mm respectively. (b) Zoomed view of upper AV leaflet highlighting the leaflet domain and LV chamber mesh element details.
Table 4 -3: Material properties of the LV-LVAD interaction model incorporating CVS and heart contraction.
Parameter Value
Blood density ( ) 1.06 × 103
Blood viscosity (Pa.s) 1.00 × 10-3
Leaflet density ( ) 1.06 × 103
Aortic leaflet Young’s 4.7 × 107 modulus (Pa)
Leaflet Poisson’s ratio 0.49
90
4.2.3.2 Mesh Generation
In this version of the simplified model, three types of COMSOL automated meshing
methods were combined: 1) an “extremely fine” free-triangular mesh was used for
the fluid domain with maximum and minimum element sizes of 0.85 mm and 0.0017
mm minimum respectively, 2) due to the important role of the upper LV chamber
wall, incorporating contractility to force the fluid (i.e. blood) through the AV, this
upper wall boundary, along with the AV leaflet edges, was discretised with 8
boundary layers made up of quadrilateral elements, and 3) a “free triangular” mesh
was employed for all remaining domains (see Figure 4 .7). Properties of the mesh for
the model are listed in Table 4 -4. The total number of triangular and quadrilateral
mesh elements was 11224 and 5394 respectively.
4.2.3.3 Material properties
For this version of the simplified model, the viscosity of blood and the elastic
properties of the valve leaflets are shown in Table 4 -3. However, for more accurate
simulation, the model dimensions were changed to represent LV dimensions, the
Young’s modulus for the AV leaflets was changed to 4.7 × 107 Pa and fluid viscosity
decreased to 1 mPa.s.
91
Table 4 -4: Mesh properties of simplified LV-pump model with CVS and heart contraction. Region of interest Mesh Properties
valve leaflet edge Number of edge elements: 192
Mean element size: 0.656 (mm)
Fluid domain Number of elements: 14494
Triangular: 10500
Quadrilateral: 3994
Mean element size: 0.822 mm
Valve leaflets thickness Number of elements: 2124
Triangular: 724
Quadrilateral: 1400
Mean element size: 0.665 mm
LV outer boundaries Number of edge elements: 360
Mean element size: 0.65 mm
4.2.3.4 Boundary Conditions
Fluid flow boundary conditions for the simplified LV-pump interaction model with
contraction included the following: a sinusoidal flow pattern with period 1 sec, mean
25 mL/s and amplitude 50 mL/s was applied at the inlet (source) boundaries, we
have used the Windkessel model to determine the pressure (Pout) according to the
equation (4.6), and was specified at the outlet (sink) boundaries and a pressure of
92
was applied at the outflow of the pump cannula, where PS is the systemic pressure of the circulation and P is the pressure differential head of the LVAD pump (see Figure 4 .8). The fixed walls of the model were set to be no-slip boundaries, and the valve leaflets were assigned a 'moving-wall' boundary condition
(Figure 4 .7), whereby the velocity of the fluid at these leaflet boundaries was set equal to the velocity of the moving wall. The stress on the valve leaflet boundaries was set to equal the fluid stress, with the leaflet root boundaries fixed.
Finally, heart wall contraction was simulated by moving the upper boundary of the ventricular chamber with a velocity according to:
(4.7)
where is the wall velocity, is the x coordinate measuring the arc length across this boundary from the left (as referenced from Figure 4.7), is the maximum velocity, and L is the total length of the upper boundary segment, equal to 5 cm. As with the valve leaflets, the fluid velocity at this boundary was set equal to the wall velocity.
93
Figure 4 .8: Windkessel model of the circulation where is the left ventricular outlet pressure, is the arterial systemic pressure, is atrial inflow to the LV, is the pump flow rate (L/min), is the blood flow ejected from the LV, is the characteristic aortic impedance, is the peripheral resistance, is the arterial systemic compliance and P is the pressure differential head of the LVAD pump.
4.2.3.5 Computational Settings
FSI simulations for this model were performed on a 3.20 GHz Intel Core i7-3930K
PC workstation, using a 64-bit Windows platform, with an applicable memory
allocation of 32 GB. The fully-meshed model exhibited approximately 65,500
degrees of freedom.
4.2.3.6 Numerical Settings
The above simplified valve-pump model with CVS and heart contraction
was implemented using a BDF integration time stepping algorithm. The followed
94
settings were used for this time-dependent solver: absolute tolerance 0.001,
relative tolerance 0.01, maximum time step 0.01s, maximum order of BDF method
set at 2, and sampling rate of saved model output set to 100 Hz.
4.2.4 Simplified LV-pump interaction model with systemic circulation, cardiac contractility and without AV
4.2.4.1 Geometry definition
Similar to the previous simplified models described, this model also utilized
the ALE finite element method to simulate the fluid/structural domain interactions
using a strong coupling method by default.
Figure 4 .9: 2D representation of the LV chamber and pump cannula without AV leaflets.
95
However, for a more simplified and computationally efficient simulation of aortic pressure (AoP) and left ventricular pressure (LVP), the geometry of this simplified model was based on the previous 2D LV-pump simplified models, however this time by excluding the AV leaflets (see Figure 4 .9). A thick homogeneous upper LV moving wall was added to implement LV contraction, with length 5.0 cm and width
0.4 cm.
Instead, AV behaviour was represented mathematically by a resistance and diode to allow flow only in one direction out of the sink boundaries representing the aorta.
Blood will flow out of these boundaries when the pressure gradient across the LV and aorta is positive, according to:
(4.8)
where Q denotes the aortic outflow, PAo denotes the aortic pressure, R is the AV resistance, and PLV is the upstream pressure within the LV chamber.
96
Figure 4 .10: (a) Fluid and structural domain meshes for the simplified LV model without AV. (b) Mesh element detail at the sink outflow boundary. (c) Zoomed view of theoutflow bound- ary highlighting the mesh element detail around the aortic outflow.
4.2.4.2 Mesh Generation
To capture sufficient detail of AV state, mesh element size was finer around the
aortic outflow boundaries than the previous simplified models (see Figure 4 .10).
Therefore, the upper LV wall moving boundary and the outflow boundary layers
were discretised with 8 boundary layers, using 1 thickness adjustment factor. For
the fluid domain, an “extremely fine” free triangular mesh was used, as listed in
Table 4 -5.
.
97
Table 4 -5: Mesh properties of simplified LV-pump model without AV leaflets.
Region of interest Simplified model without AV leaflets
LV outer boundaries Number of edge elements: 301
Mean element size: 0.80 (mm)
Outflow inner boundaries Number of edge elements: 356
Mean element size: 0.089 (mm)
Fluid structure Number of elements: 24213
Triangular: 19483
Quadrilateral: 4730
Mean element size: 0.77 (mm)
Myocardial internal boundaries Number of edge elements: 31
Mean element size: 0.7 (mm)
4.2.4.3 Material properties
The viscosity of blood was similar to the previous model (see Table 4 -3). The
moving myocardial wall was modelled as being linear elastic, with properties as
shown in Table 4 -6.
98
Table 4 -6: Material properties of the myocardium in simplified LV model without AV
Parameter Value
Wall density ( ) 1.06 × 103
Myocardial Young’s 1.0 × 107
modulus (Pa)
Myocardial Poisson’s ratio 0.49
4.2.4.4 Model equations
4.2.4.4.1 Windkessel model
To simulate the systemic circulation, a simple Windkessel model was employed,
characterized by the following equations (see Figure 4 .11):
(4.9)
(4.10)
(4.11)
where P is the aortic pressure, is the left ventricular outlet pressure, is the
arterial systemic pressure, is the pump flow rate (L/min), is the blood flow
99
ejected from the LV, is the characteristic aortic impedance, is the peripheral
resistance and is the arterial systemic compliance.
Figure 4 .11: Windkessel model used with simplified LV-pump model without AV leaflets.
4.2.4.4.2 Pump model
The differential pressure head (ΔP) across the pump outlet was modeled using three
equations as reported by Lim et al. [149]; the motor windings electrical equation
(4.12), the electromagnetic torque transfer equation (4.16), and the pump hydraulic
equation linear equation (4.17).
100
i) Motor windings electrical equation
(4.12)
where V is the motor terminal voltage, I is the motor current, is motor winding
resistance (1.38 ) and is the motor winding reactance. is the back
electromotive force (BEMF) given by:
(4.13)
where = 8.48 and is the motor electrical speed ( , where is
the impeller speed in rad/s). Due to the synchronization between BEMF and motor
electrical current to produce maximum torque efficiency, equation (4.12) can be
written as:
(4.14)
where L = 0.439 is the motor winding inductance. V was determined using a
proportional controller to track the desired pump speed according to:
(4.15)
where K is constant (1 V.s/rad) and is the pump speed set point. ii) Electromagnetic torque transfer equation
(4.16)
101
Where Te is the output electromagnetic torque and J is the moment of inertia of the
-6 2 impeller (7.74×10 kg/m ). The coefficients c1, c2, c3 and c4 are viscosity-
4 dependent parameters, set to constant values of c1=1576.8 kgs/m , c2=0.0119
-5 2 -10 2 kgs/m, c3=1.92×10 m kg/s, and c4=3.14×10 m kg/s.
iii) Pump hydraulic equation
(4.17)
where c5, c6, c0 are viscosity-dependent parameters, fixed to values of c5= -32343
10 5 kg/(m s ), c6=0.25331kgs/m, and c0= -39.463 Pa.
4.2.4.5 Boundary Conditions
To simulate a periodic diastolic inflow, a sinusoidal flow pattern with period 1 sec,
mean flow rate 40 mL/s and amplitude 80 mL/s was applied at the inlet (source)
boundaries. Furthermore, we have used the Windkessel model to determine the
pressure (Pout) according to the equation (4.6), and this pressure was specified at the
outlet (sink) boundaries, a pressure of was applied at the
outflow of the pump cannula, where the Lf is the blood inertance, and the fixed walls
of the model were set to be no-slip boundaries, as shown in Figure 4 .11. To simulate
LV contraction, an external force was applied to the upper LV moving boundary
given by Equation 4.18:
(4.18)
102
where F is a periodic forcing function, and Fmax is the value of maximum
force when the wall is fully contracted.
4.2.4.6 Computational Settings
The FSI simulations were performed on a 3.20 GHz Intel Core i7-3930K PC
workstation, using a 64-bit Windows platform, with an applicable memory
allocation of 32 GB. The fully-meshed model exhibited approximately 76,819
degrees of freedom.
4.2.4.7 Numerical Settings
The realistic LV-pump model was implemented using the following numerical
integration settings: absolute tolerance 0.001, relative tolerance 0.01, maximum time
step 0.01s, maximum BDF order set to 2, and sampling rate of saved model output set
to 100 Hz.
4.2.5 Realistic LV-Pump geometry model with and without AV leaflets
4.2.5.1 Computational approach and geometry definition
This section outlines the scheme developed for simulating aortic valve function
inside the left ventricle and under LVAD support using a more realistic LV
103
geometric shape. This model characterizes a 2D section through the left side of the
heart, lying in a long axis plane passing through the apex of the heart, bisecting the
mitral valve leaflets, and passing through the aortic outlet flow tract. Our purpose
has been to change the model’s dimensions and parameters so it models a human
heart. The model consists of the left ventricle of height 7 cm from base to the apex,
with a diameter of 5.18 cm and a varying wall thickness of 0.8 to 0.95 cm (see Fig.
4.11). The model also includes narrow curved structures representing the AV
leaflets. The LV boundaries consisted of an endocardial (inner) border lining the
cavity of LV, and an epicardial (outer) border, as shown in Figure 4 .12.
Figure 4 .12: 2D geometry of the LV and AV (long axis plane) with LVAD cannula attached at the apex. The diameter of the LVAD cannula was 0.87 cm. 104
Key model parameters including blood density, blood viscosity, leaflet density, aortic leaflet Young’s modulus, leaflet Poisson ratio and myocardium properties are shown in Table 4 -7 and Table 4 -8. The ALE FSI method was used to simulate the dynamics of the aortic valve and the LV wall. The shape of the aortic root was based on the idealized geometric description of De Heart et al. [18]. The AV was simplified with an outflow region of diameter of 2.1cm, consisting of two flexible curved shape leaflets with equal length of 1.25 cm and thickness 1 mm. The myocardium wall thickness of around 0.90 cm was modified from Bogaert et al.[150].
A thin-walled cannula of width 0.87 cm and length 0.8 cm, representing the LVAD, was inserted into the ventricular chamber wall. The dimensions of all these structures were adapted from McQueen et al. [109] and were arranged so that the geometry of the 2D left ventricle closely matched echocardiographic data of the normal human heart. However, this LV-pump model was also modelled using version without the AV leaflets, to produce a more computationally-efficient model, as shown in Figure 4 .13. The ALE Navier-Stokes equations [151] were used to model the blood inflow into the LV on boundaries placed at the left end of the channel (labelled as "source" in Figure 4 .13), whilst aortic outflow was modelled as a varying pressure boundary condition (labelled "sink" in Figure 4 .13). Outward flow at the "sink" boundaries was simulated using a Windkessel model of the circulation, connecting aortic pressure and flow. These equations governing the
Windkessel circuit were implemented as global differential equations in the 105
COMSOL solver. Heart wall contraction on the outer boundaries of the ventricular
chamber was also incorporated by applying an external periodic sinusoidal load
profile on the LV epicardial boundaries.
Figure 4 .13.Two-dimensional geometry of the LV (long axis plane), LA, and LVAD cannula without the AV leaflets.
4.2.5.2 Material properties
In this model, the viscosity of blood and the elastic properties of the valve leaflets
were obtained from the literature to represent healthy human values [144, 145]. The
main properties of blood used in this simulation are shown in Table 4-7. 106
In addition, the ventricular wall was modelled as being linear elastic with appropriate Young’s modulus and Poisson’s ratio properties, are shown in Table 4-8.
Table 4 -7. Material properties of 2D realistic LV model
Parameter Value
Blood density ( ) 1.06 × 103
Blood viscosity (Pa.s) 1.00 × 10-3
Leaflet density ( ) 1.06 × 103
Aortic leaflet Young’s 4.2 × 107
modulus (Pa)
Leaflet Poisson’s ratio 0.49
Table 4 -8. Material properties of the myocardium
Parameter Value
Wall density ( ) 1.06 × 103
Young’s modulus (Pa) 1.0 × 107
Poisson’s ratio 0.49
107
4.2.5.3 Mesh Generation and Quality
A fine mesh resolution was used for the fluid domain and AV leaflet boundaries, as
illustrated in Figure 4 .14 and Figure 4 .15. Figure 4 .16 shows the element quality (q)
over the whole model in the range 0 to 1. The mean element quality of the triangular
and quadrilateral elements in the fluid domain, myocardial walls and AV leaflets
were 0.91, 0.93 and 0.88 respectively (Table 4 -9), which were significantly greater
than the q thresholds known to affect solution accuracy [85].
Figure 4 .14: (a) 2D realistic LV model mesh. (b) Zoomed-in view of boundary mesh layers at the interface between the LV walls and fluid. (c) Zoomed-in view of AV leaflet tips highlighting the mesh element detail.
However, to obtain accurate simulation results, we utilized three types of automated
meshing methods as provided by COMSOL: (1) a “fine” resolution for the fluid
domain with maximum and minimum element sizes of 3.22mm and 0.092mm
respectively, (2) a boundary layer mesh was used for all fluid-LV cavity boundaries
with two boundary layers and 6 thickness adjustment factors, and (3) “free 108
triangular” meshing was used for the remaining domains (Figure 4 .14). The total number of mesh elements in the fluid domain was 17777 (see Table 4 -9).
Table 4 -9. Mesh properties of realistic LV model.
Region of interest Realistic LV model
Valve leaflet edge Number of edge elements: 398
Mean element size: 1.16 (mm)
Valve leaflet tips Number of edge elements: 87
Mean element size: 0.084 (mm)
Myocardial wall edges Mean element size: 2.44(mm)
Number of edge elements: 1237
Mean element size: 2.5 (mm)
Myocardial thickness Number of elements: 663
Mean element size: 3.00(mm)
Average element quality: 0.93
Fluid domain Number of elements: 17777
Triangular: 17117
Quadrilateral: 660
Mean element size: 0.822(mm)
Average element quality:0.91
Valve leaflets thickness Number of elements: 2218
Triangular: 2218
Quadrilateral: 0
Mean element size: 1.23(mm)
Average element quality: 0.91
109
Figure 4 .15: Mesh element size in 2D realistic LV model showing the aorta (Ao) and AV leaflet regions. (a) Boundary layer sizes at the Ao (b) zoomed view of the AV leaflet tips highlighting the element sizes in mm.
Figure 4 .16: (a) Element quality of realistic LV model (q) in the range from 0 to 1. (b) Zoomed-in view of boundary element quality at the interface between the LV wall and fluid. (c) Zoomed-in view of AV leaflet tips, highlighting the element quality around the AV.
110
During movement of the heart wall and the AV leaflets, adjacent triangular elements
will deform in accordance to COMSOL's in-built moving mesh interface, so that the
mesh quality remains as high as possible. If however the triangles do degenerate,
there is an option in the interface to perform a remesh if the overall mesh quality
falls below a threshold value. This option, however, was not required in the
simulations of this section, but could be implemented if larger deformations were
imposed onto the model.
4.2.5.4 Model Equations 4.2.5.4.1 LV model
Pulsatile flow blood was characterized as being laminar, Newtonian, viscous and
incompressible. The fluid was described by the Navier-Stokes equations for
incompressible flow:
where is the fluid density, is the viscosity, u is the velocity of the fluid and p is the pressure.
The fluid flowed into the ventricle from the right source (inlet) boundaries (see
Figure 4.12). At these entrance boundaries, the flow is assumed to have fully
developed a laminar profile. The valves were modelled as an isotropic Hookean
elastic solid, with constitutive law expressed with Einsteinian indicial notation as:
111
(4.20)
where E is the Young’s modulus, is Poisson’s ratio, are the Cauchy strain and
stress tensors respectively, and represents the Kronecker-delta tensor:
This modelling was carried out automatically using the Parallel Direct Sparse Solver
(PARDISO) solver in COMSOL using the FSI module [152].
4.2.5.4.2 Windkessel model of the CVS
To simulate the systemic circulation, a simple Windkessel model was employed,
characterized by the following equations (see Figure 4 .17):
(4.21)
(4.22)
(4.23)
where P is the aortic pressure, is the left ventricular outlet pressure, is the
arterial systemic pressure, is the pump flow rate, is the blood flow ejected
from the LV, is the characteristic aortic impedance, is the peripheral
resistance and is the arterial systemic compliance. 112
The fluid density and viscosity are listed in Table 4 -10. These values correspond to the physical properties of blood in the large arteries and heart cavities as reported by
Yilmaz et al. [153].
Figure 4 .17: 2D realistic representation of the LV, AV, pump cannula and Windkessel model of the circulation.
113
Figure 4 .18: Two-dimensional realistic representation of the LV, pump cannula and Windkessel model of the circulation without the LV leaflets.
In the realistic version of the LV-pump model without AV leaflets ( Figure 4 .18), the aortic valve was represented by a resistance (Rout) and a diode (Dout) to allow flow only in one direction from the sink boundaries, representing flow in the aorta. Blood flow across the aortic valve was non-zero only when the pressure gradient across the
LV and aorta was positive:
114
(4.24)
where Q is the blood flow through the AV, R is the resistance of the aortic valve, P1
the upstream LV pressure, and P2 the downstream aortic pressure.
Table 4 -10 Fluid material properties in 2D realistic LV model.
Density (kg/m3) viscosity (mPa.s)
1060 1
4.2.5.4.3 Pump model
4.2.5.4.3.1 Average pump flow and differential pressure estimation model
Due to the insensitivity of IRBPs to preload, overpumping or underpumping
conditions, which can potentially endanger implant recipients, such scenarios can
readily occur if pump control is not properly implemented. One design goal of an
IRBP controller is to be able to accurately and reliably predict pump flow as well as
differential pressure without the need for additional implantable sensors. Additional
complications such as residual ventricular function dependent on the amount of
residual contractility, as well as insufficient venous return, may also influence the
occurrence of undesirable conditions [136].
115
The differential pressure head (ΔP) across the pump outlet was therefore modelled using three equations, as reported by Lim et al. [149]: (1) the motor windings electrical equation (4.25), (2) the electromagnetic torque transfer equation (4.29), and (3) the pump hydraulic linear equation (4.30).
i) Motor windings electrical equation
(4.25) where V is the motor terminal voltage, I is the motor current, is motor winding resistance (1.38 ) and is the motor winding reactance. is the back electromotive force (BEMF) given by:
(4.26) where = 8.48 and is the electrical speed ( , where is the impeller speed in rad/s). Due to the synchronization between BEMF and motor electrical current to produce maximum torque efficiency, equation (4.25) can be written as:
(4.27)
where L = 0.439 is the motor winding inductance. V was determined using a proportional controller to track the desired pump speed according to:
116
(4.28)
where K is constant (1 V.s/rad) and is the pump speed set point.
ii) Electromagnetic torque transfer equation
(4.29)
where Te is the output electromagnetic torque and J is the moment of inertia of the
-6 2 impeller (7.74×10 kg/m ). The coefficients c1, c2, c3 and c4 are viscosity-
dependent parameters fixed to values of c1, c2, c3, and c4.
iii) Pump hydraulic equation
(4.30)
where c5, c6, c0 are viscosity-dependent parameters, fixed to values of c5, c6, c0.
4.2.5.5 Boundary Conditions
Fluid flow boundary conditions of the model included the following: a sinusoidal
flow pattern with period 1 sec, mean flow rate 40 mL/s and amplitude 80 mL/s was
applied at the inlet (source) boundaries, we have used the Windkessel model to
determine the pressure (Pout) and was specified at the outlet (sink) boundaries, a
117
pressure of was applied at the outflow of the pump cannula,
where the Lf is the blood inertance, the fixed walls of the model were set to be no-
slip boundaries, the valve leaflets were assigned a 'moving-wall' boundary condition,
as shown in Figure 4 .19, and LV contraction was implemented by applying a load on
the endocardium of the septum and the posterior left ventricular wall according to
Equation 4.31, which will be detailed section 4.2.6. The stress on the valve leaflet
boundaries was set to equal the fluid stress.
Figure 4 .19: Boundary conditions of realistic LV-pump model.
4.2.5.6 Computational settings
118
FSI simulations were performed using COMSOL finite element software on a
Windows 64-bit platform with 3.20GHz processor employing an Intel Core i7-3930K
PC workstation, with an applicable memory allocation of 32 GB.
4.2.5.7 Numerical Settings
The realistic LV-pump model was implemented using the following numerical
integration settings: absolute tolerance 0.001, relative tolerance 0.01, maximum
integration time step 0.01s, maximum BDF order set to 2, and sampling rate of saved
model output set to 100 Hz.
4.2.6 Simulating LV and AV dynamics
To model the dynamic contraction and relaxation of the ventricle, the myocardial
and septal walls were modelled as an incompressible elastic isotropic materials with
properties shown in Table 4 -11.
Table 4 -11 Material properties for the solid model.
Density (kg/m3) Young’s Poisson ratio modulus (MPa)
LV wall 1040 100 0.49
AV leaflets 1060 470 0.49
For the LV wall structural mechanics computations, a time- and coordinate-
dependent load was applied at the LV outer boundaries (i.e. myocardial and
interventricular septal walls) according to:
(4.31) 119
where F is the force per unit area, and Fmax is the maximum force when the walls are
fully contracted.
In this method, the incorporation of dynamic heart wall motion into simulations
plays an important role in modelling the open/closed state of the AV, thus regulating
flow during the cardiac cycle. The fluid boundary conditions at these contacting
regions was set at moving-wall. The apex outer walls were set to be free-wall, and
all sides of the cannula were set to be fixed (i.e. zero displacement).
4.2.7 Investigation of Cardiovascular Interaction with a Left ventricular Assist Device 4.2.7.1 Speed and Parameter Variation studies
Numerical models, able to simulate CVS response in the presence of an implantable
rotary blood pump (IRBP), have been widely used as predictive tools to investigate
the interaction between the CVS and the LVAD under various operating conditions.
Contractility, ventricular preload, and afterload, which may endanger patients, can
be investigated individually using a numerical model. In addition, such models can
provide additional understanding of the dynamics of the assisted circulation, sheding
light on research questions not easily answered in vivo. Specifically, such models
can help investigate the impact of circulatory perturbations on the haemodynamics
of the heart.
120
This section describes an extension of the previous method sections to investigate
LV-pump interaction over a wide range of operating conditions, including
alterations in systemic vascular resistance and total blood volume by varying the rate
of inlet flow. These variations are important tools in testing the interaction between
the heart and the LVAD under different operating scenarios and to avoid
unacceptable dangers to recipients if pump control is not appropriately implemented.
4.2.7.2 Simulation Protocols
In this set of numerical experiments, model parameters illustrated in Table 4 -12,
which correspond to the Healthy condition (H1), were used as starting point values
for a normal subject. To allow simulation of various physiological conditions, two
cardiovascular parameters were altered: afterload and preload.
4.2.7.3 Effect of speed ramp
The model was tested under one physiological “Healthy condition” (H1), with syst-
em parameters set at baseline values. The simulation was first carried out using a
constant reference pump speed input, starting at 50 rad/s at t=0 s. During this
test, the impeller speed set point ωset was steadily increased from 50 rad/s to 200 rad/
s in 50 rad/s steps, then increased from 150 rad/s to 200 rad/s in 30 rad/s steps, in o-
rder to cover the full range of pumping states. The physiological signals, instantane-
121
ous pump impeller speed (ω) and motor current (I), were post-processed and saved
from the pump model.
4.2.7.4 Effect of parameter variations
In order to evaluate the response of the AV for different cardiovascular states, the m-
odel was tested under two sets of parameters for a wide range of pump operating set-
tings:
i. varying SVR.
ii. changing the total blood volume Vtotal, (see Table 4 -12 and Table 4 -13).
These settings were evaluated using both the simplified and realistic versions of the
2D LV-LVAD models.
Table 4 -12. Simplified LV-pump model parameters for the healthy and abnormal condition subjects: Systemic peripheral resistance, Rsa; and total blood volume, Vtotal.
Variable H1 H2 H3
-1 Rsa (mmHg.mL ) 0.74 0.74 0.40
Vtotal (mL) 5600 3400 5600
First, the “Volume low condition” (H2) (or preload), was simulated for both simplif-
ied and realistic LV-LVAD models by varying the blood volume input and the pump
speed: Vtotal was decreased to 3400mL and 1875ml respectively, and the simulation
was continued for 6 seconds to allow the system to reach steady state. 122
Next, an “Afterload condition” (H3) was simulated. In this test, the SVR parameter
was decreased by 50% for both models. This H3 simulation was conducted to deter
mine whether the model, in combination with alteration of LV parameters, was able
to provide normal AV open-close functionality during afterload decrease.
Table 4 -13: Realistic 2D LV-LVAD model parameters for healthy and abnormal conditions: Systemic peripheral resistance, Rsa; and total blood volume, Vtotal.
Variable H1 H2 H3
-1 Rsa (mmHg.mL ) 0.74 0.74 0.40
Vtotal (mL) 5600 1875 5600
123
Part III-Results and Discussion
124
5 Chapter 5 - Fluid-Structure Interaction in a Simple Model of an Assisted Left Ventricle
This chapter presents the first steps in simulating the effects of outlet pump pressure
on aortic valve closure of the heart assisted by an implantable blood pump. A 2D
fluid structure interaction aortic valve model is presented with blood flow in the left
ventricular chamber using the Arbitrary Lagrangian–Eulerian finite element
formulation to predict AV closure during outflow of blood from the left ventricle
into the left ventricular assist device (LVAD).
5.1 Introduction
Heart disease represents one of the biggest causes of death in the world [2]. Due to
the shortage in donor hearts, an artificial heart can be a bridge to transplantation or
serve as a destination therapy for patients with heart failure. It is well known that the
changes in hemodynamic behaviour of the heart can alter the physiological structure
of heart valves during LVAD support [154]. There are various imaging tools that can
visualize mechanical characteristics of the left ventricle (LV) as well as blood flow
inside the ventricle, qualitatively, but they do not provide a complete picture of the
status of heart valves during LVAD support following implantation [155-157].
125
FSI simulations have been extensively used for the purposes of heart function
assessment and heart disease diagnosis. A model of human left ventricular-pump
interaction by the finite element method was developed to simulate the
haemodynamics inside the assisted LV. The model was formulated with dimensions
and parameters to mimic simple 2D AV dynamics and LVAD behaviour. The
primary focus of this chapter is to present results using a simplified model of the LV
chamber to examine the effect of intra-ventricular flow under the action of an assist
device on AV status. The simulations of this chapter provide a first step in
investigating the effect of outlet pump pressure on AV closure. This will potentially
provide significant insights into LV function under LVAD assistance.
The main results of this chapter were published in Alonazi et al. [158].
5.2 Simplified LV-pump model
5.2.1 Results:
Simulations using the simplified LV-pump model revealed that model behaviour
could effectively be separated into two phases: aortic valve closed and open, as
shown from the simulated outlet pump pressure result of Figure 5 .1 , which has
been calculated from the cannula inlet pressure minus the differential pressure head.
The AV closed phase was characterized by increasing pressure in the LV during the
early stage of filling. Moreover, the LVAD pumping action kept the LV pressure
126
lower. However, during the second phase, this pressure dropped from its peak when the aortic valve opened.
Figure 5 .1: Simulated pressure at pump outlet over a single 1 s cardiac cycle, where ' * ' and ' ● ' indicate AV closing and opening times, respectively. Label 'c' re- presents the period during which the AV was closed, and 'o' is the period in wh- ichthe valve was open.
From Figure 5.1, we see that a low blood velocity occurs at the source boundaries during the closing phase (t=0.4s), rising sharply and accelerating while the aortic valve was in its closed state (t=0.4-0.675s), reaching its maximum value of 0.25m/s at t=0.675 s.
It can be observed that during this period, blood moves toward the LVAD inlet due to suction created by the pump. Meanwhile, the pressure at the pump outlet increases before reaching its maximum value of 35.0 Pa at t=0.675s, as illustrated in Figure
5 .1. In this phase, the aortic valve is closed, however during the AV opening phase 127
starting at t=0.7s, the pump pressure starts to increase gradually, reaching its
maximum value of 185.8 Pa at t= 1s. The valve leaflets, as observed from Figure 5 .2,
are fully opened at t=1s, before starting to close again. The leaflets are fully closed
by t=1.56s resulting in an open phase duration of 0.86s.
Figure 5 .2: Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. (a) Aortic valve at closing phase. (b) Aortic valve at opening phase.
5.2.2 Discussion We have presented results from a two-dimensional FSI ALE model of aortic valve
motion in a simplified LVAD-assisted LV chamber. The simulations confirmed that
when the AV opens, pressure falls at the LVAD inlet, similar to the findings obtained
by Shi et al. [135]. However in their model, they did not incorporate any 128
computational fluid dynamics to determine pressure and velocity distributions during
LVAD support, ignoring the effects of aortic valve interaction between the blood and
moving solid structures.
This simulated pressure variation during AV movement, particularly during its
closure phase, provides significant insight into LV function under LVAD assistance,
offering the potential of improving the accuracy of current LVAD control systems to
ensure patient safety and comfort.
5.3 Simplified LV-pump model with systemic circulation
5.3.1 Results:
In this section, results are presented from a 2D LV chamber model with CVS,
capable of simulating aortic valve dynamics (closure/opening) under LVAD support,
which can be useful in the design and evaluation of physiological pump control
algorithms to ensure patient safety and comfort, as well as evaluating aortic valve
pathologies.
As in the previous results, model behaviour could effectively be divided into two
phases: aortic valve closed and open, as shown from the outlet pump pressure in
Figure 5 .3. AV open/close state was characterized from two simulated signals: the
distance between the valve leaflets and the outlet pump pressure during the early 129
stage of filling. During the first closed phase, the pumping action of the LVAD kept
the AV pressures low. During the second phase, this pressure dropped from its peak
value when the aortic valve opened.
From Figure 5 .3(b) a low fluid velocity is observed at the source boundaries during
the closing phase (t=1.51s), rising sharply and accelerating whilst the aortic valve is
still in its closed state (t=1.51-1.88s), reaching a maximum value of 4m/s at t=2.00 s
when the valve has fully opened.
Figure 5 .3: (a) Simulated pressure from the pump outlet over a single, 1 s cardiac cycle, where ' * ' and ' ● ' indicate aortic valve opening and closing times, respectively. ‘c’ represents the period during which the aortic valve is closed, and ‘o’ is the period in which the aortic valve is open. (b) Zoomed-in view of pumpoutlet pressure. T- he pressure shown at the position of the arrows increases transiently on aortic valve closure.
130
It can be observed that during the first phase, blood moves toward the pump inlet due to the suction created by the pump and closure of the aortic valve. Meanwhile, the pressure at the pump outlet shown at the position of the arrows in Figure 5.3 (i.e. at the instant of AV closure) reveals a transient increase on aortic valve closure, before reaching its maximum negative amplitude of -7000 Pa at t=1.45s, as illustrated in
Figure 5 .3(b). After this instant, the aortic valve remains closed until the AV opens at t=1.88s. After this opening, the pump pressure begins to decrease gradually (Figure
5 .3 (b)). The aortic leaflets, as observed from Figure 5 .4, are fully opened at t=2.03s before starting to close again. The leaflets are fully closed by t=2.44s resulting in an open phase duration of 0.56s.
131
Figure 5 .4: Snapshots of computed LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. (a) Aortic valve in the closed phase. (b) Aortic valve in the open phase.
132
5.3.2 Discussion
Development of valve abnormalities after LVAD placement is common among
patients with advanced heart failure, likely due to the fact that the LVAD induces
different haemodynamics by changing the direction of blood flow from the apex of
the heart, largely bypassing the LV, directly to the aorta. This abnormal recirculation
modifies the pressure and stress on the AV leaflets, leading to remodelling of the
valve [159]. In particular, aortic valve functional problems have been reported by
Rose et al. [2] and Letsou et al. [3], who note that that aortic insufficiency (AI) and
Aortic stenosis (AS) are more prevalent in LVAD patients. In addition, Park et al.
[48] reported that within 6–12 months of LVAD implantation, 50% of patients with
pulsatile LVADs developed AI or AS. A major consequence of LVADs is excessive
flow bypassing the AV, resulting in diminished and infrequent AV opening which
could also extend to the systolic phase [5].
Managing native aortic valve insufficiency during LVAD placement is very
challenging. A variety of procedures to prevent aortic insufficiency have been
reported, including aortic valve replacement with a bioprosthetic valve, or primary
closure of the aortic valve opening [160-162]. However, these surgical techniques
could cause further complications, impairing the LVAD as being a viable alternative
to transplantation [163, 164]. Currently, MRI and Echocardiography are the only
techniques used clinically to evaluate aortic valve abnormal pathophysiology in
LVAD recipients in real-time [52, 165].
139
Alternatively, computational modelling of blood flow has been used extensively to study LV haemodynamics. Several FEM models of the left ventricle and aortic valve have been developed to simulate leaflet motion due to blood flow. Early studies of heart valve FSI models include the work of Peskin [110], who used an immersed boundary method to describe a 2D model of the left heart. Watanabe et al. [166] addressed the limitations of the Peskin approach, using instead the ALE approach to simulate cardiac mechanics during cardiovascular flow. Alternately, De Hart et al.
[18, 127] used a fictitious domain (FD) method to simulate closure of a flexible heart valve during systole. Loon et al. [130], extended this approach by linking fluid dynamics with adaptive meshing for the flexible leaflet. Stijnen et al. [131] also simulated the dynamic behaviour of a two-dimensional moving rigid heart valve using the FD method. Lastly, McCormick et al. formulated an FSI model of the left ventricle with LVAD using a modification of the Newton–Raphson/line algorithm and optimizing the interpolation scheme at the fluid–solid boundaries [167, 168].
However, these models did not include simulation of AV closure during LVAD support. Understanding how the AV behaves under LVAD support may allow the detection of the open/closed state of the valve from LVAD pressure/flow sensor transducers alone, paving the way for more sophisticated pump control algorithms which take into account the AV state.
In this chapter, a simplified 2D FSI model of the aortic valve and ventricle during
LVAD support has been presented, including the systemic circulation. The results
140
confirm that when the AV opens, less blood flows through the LVAD compared to
when the valve is closed, similar to the results reported by Shi et al. [135].
However, our simulated pressure variations during AV movement, particularly
during its closing phase, provide significant insight into LV function under LVAD
assistance, offering the potential of improving current LVAD control systems to
ensure patient safety and comfort.
5.4 Conclusion
Previous FSI models of ventricular function with LVAD support have ignored the
impact of AV closure on blood dynamics in the assisted left ventricle, as presented in
this chapter. Blood flow in a 2D model of the left ventricle with aortic valve was
simulated using the finite element model. This model can be used to investigate heart
aortic valve closure during LVAD support. The next chapter includes results from
modelling a realistic ventricular geometry, as well as adding the contractile activity
of the LV in order to simulate an entire cardiac cycle. To date, few modeling studies
which have considered valve-blood interaction in addition to ventricular wall-blood-
interaction have incorporated LVAD pumping action.
141
6 Chapter 6 - Simulation of Aortic Valve Response during Ventricular Assist Device Support
In this chapter, analysis of simulated LVAD motor current waveform was
performing to derive useful parameters for evaluating aortic valve status. A 2D fluid
structure interaction model of the LV and aortic valve in the presence of LVAD flow
are presented to predict AV closure during LVAD outflow of blood to derive an
automatic pump speed control method.
6.1 Introduction
Congestive heart failure (CHF) is a serious health condition characterized by the
inability of the heart to supply sufficient blood flow to tissues and organs in the body.
Left ventricular support with pump devices has been an essential element in cardiac
health care for several decades. It is therefore important to understand the interaction
between the cardiovascular system and a cardiac pump device. Furthermore, to avoid
valvular stenosis and thrombogenesis reported in previous investigations [49, 169],
the monitoring of aortic valve opening and closing is important during pump support.
This chapter provides preliminary steps in simulating the role of outlet pump pressure
on aortic valve function of the heart assisted by an implantable blood pump. A simple
2D FSI aortic valve model is presented with blood flow in the left ventricular
chamber using the ALE Finite Element Method to predict AV closure during blood
outflow from a left ventricular assist device (LVAD).
142
The aim here is to:
● Simulate the response of the aortic valve in the presence of an LVAD in the 2D
simplified and realistic LV geometries.
Simulate the response of the models over a range of pump operating points (i.e. over
a wider pump speed range).
AV state was evaluated using the simulated LVAD motor current waveform as an
index, using two 2D simplified computational models of AV dynamics
(closure/opening) under heart pump support. Such models will be useful in the design
and evaluation of physiological heart pump control algorithms to achieve the balance
between LVAD pumping and mechanical circulatory support against disruption to
AV movement, which may lead to various AV pathologies. Therefore, the objective
of this chapter is to investigate how the motor electrical current profile changes with
AV state and pump speed. Identifying the AV state according to the non-invasive
motor current signal could serve as a useful an input to a pump speed controller to
prevent highly negative pressures from occuring in the left ventricle leading to wall
suction, as well as ensuring the AV properly opens and closes over the complete
cardiac cycle.
The main results of this chapter were published in Alonazi et al. [170].
6.2 Simplified LV-pump model
Model behaviour could effectively be divided into two phases: aortic valve closed and open, as shown from pump motor current in
Figure 6 .1. The valve ejection phase (VE) was characterized by two waveforms: the
pump motor current and the pump
143
Figure 6 .1: Simulated motor electric current and pump impeller speed at two motor speed set points (100 and 150 rad/s), where ' ■ ' and ' ● ' indicate AV opening and closing times, respectively, (a)
Simulations using a motor speed set point of set =100 rad/s. Periods 'O' and 'C' represent the phases during which the AV is open and valve closed respectively. (b). Simulations using a motor speed set point of set =150 rad/s. Periods 'O' and 'C' represent the phases during which the AV is open and closed, respectively. Max PCT and Min PCT are the maximum and minimum values of current threshold during AV closure, respectively. The motor current waveform at Max PCT begins to decrease more rapidly once the aortic valve is closed at the closing notch (CN).
144
speed. In each simulation, the impeller speed set point was increased from 50 rad/s to
200 rad/s in 50 rad/s increments, in order to cover the full range of aortic valve state transitions (from ventricular ejection (VE) to aortic valve closed (VC)). The physiological signals, instantaneous pump impeller speed (ω) and motor current (I), were simulated from the pump model equations.
Figure 6 .1 shows the waveforms obtained at two different speed set points of 100 rad/s and 150 rad/s, illustrating the relationship between the peak motor current and instantaneous pump speed in the 2D model, indicating an inverse correlation of motor current with pump speed.Transition from state VE to state VC occurred with increasing pump speed. However, at a pump speed set point of 200 rad/s, VC state was continuously maintained, whereby the AV remained closed throughout the entire cardiac cycle (ANO) with no blood flow to the proximal aorta. Furthermore, a peak motor current (MAX PCT) of 0.203A was observed during the AV closing phase
(t=3.15s) for 100 rad/s, decreasing rapidly throughout the AV closed state
(t=3.15-3.85s), reaching a minimum value (MIN PCT) of 0.1067A at t=3.49s.
Figure 6 .2 shows the simulation results of AV movement during LVAD support, where the opening time at t=2.85 s was approximately the same for both speed set points. However, the closure time was delayed by 10 ms to t = 3.16s at the motor speed set point of 150 rad/s. During AV opening, there were also small oscillations in the current waveform, probably due to mechanical flutter of the valve leaflets.
145
Figure 6 .2: Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. (a) Aortic valve and blood motion for = 100 rad/s. (b) Aortic valve and blood movement for = 150 rad/s
146
6.3 Realistic LV-Pump model with AV leaflets
In the section, we present results of AV dynamics simulated from a 2D realistic LV
computational model under heart pump support. AV state is examined from non-
invasive simulated LVAD motor current as an index. The model could serve as a
useful tool in the design and evaluation of physiological pump control algorithms to
achieve a balance between LVAD speed and disruption to AV movement which can
lead to various AV pathologies. Therefore, the objective of these simulations was to
investigate the correlation between AV status and the motor electrical current
waveform using the 2D LV-pump interaction model. Also, identification of AV state
according to motor current could provide an input to a pump speed controller to
prevent highly negative pressures in the left ventricle leading to wall suction, as well
as ensuring the AV opens and closes over the cardiac cycle.
The hypothesis examined with this set of simulations is that by using the non-
invasive measures of instantaneous pump electrical current and impeller speed, it is
possible to detect various AV states including:
Open AV state during Ventricular ejection (VE)
Closed AV state during diastole (AC)
AV remaining closed over the entire cardiac cycle (ANO).
AV regurgitation (AR), which frequently occurs in LVAD heart failure
patients. 147
6.3.1 Identification of aortic valve states for LV- pump interaction models with AV
In order to assess whether pump speed and motor current are potential indicators of
aortic valve state transitions, it is necessary to independently verify the physiological
state of the aortic valve in the simulations. This was achieved by using the pump int-
rinsic parameters indicated in Table 6-1. Model behaviour can effectively be divided
into two phases: systole and diastole. An initial inspection of the distance between l-
eaflets and the difference between left ventricular pressure (LVP) and Aortic pressu-
re (AoP), indicated the presence of two physiologically significant aortic valve state-
s (i.e. open and closed) (see Figure 6.4).
Table 6 -1 A summary of physiological and pump basic parameters identifying AV state.
Physiological Pump Basic Parameters Parameters AV State Motor Pump Distance between AV LVP current speed leaflets
AC decreasing VE increasing >AoP decreasing > 0 148 Figure 6 .3: Simulated left ventricular pressure (sold black line) and aortic pressure (dotted red line) at a low pump speed set point of set = 50 rad/s. The method here is that by using only the non-invasive measure of instantaneous pump current to evaluate AV dynamics during LVAD support, it is possible to detect a range of AV states according to these classifications: 1. In the systolic phase, due to the positive difference between left ventricular pressure and aortic pressure (LVP > AoP), (see Figure 6 .3), blood flows through the AV from the ventricle to the aorta causing the AV to open. During this state, LVAD pump flow is also increased. Following LV contraction, the pump differential pressure decreases, increasing the pump flow and torque on the impeller, causing the impeller speed to fall and the motor current to increase (see Figure). 2. During isovolumic relaxation, left ventricular pressure (LVP) decreases, causing the aortic valve to close, (LVP < AoP), increasing the differential pressure across the pump, causing the pump flow to fall. Subsequently, the torque on the impeller decreases, causing impeller speed to increase and motor current to decrease. 149 3. If the aortic valve remains closed throughout the entire cardiac cycle (ANO), the aortic flow (Qav) will equal zero at all times. This condition could not be maintained in our model, due to the difficulty in simulating permanently contacts AV leaflets.. 4. In the aortic valve regurgitation state (AR), LV pressure is excessively low due to high pump speed, aortic root pressure is always higher than the left ventricular pressure (AoP>LVP) as a result of LVAD unloading. This occurs during diastole. 6.3.2 Detection of AV State The realistic 2D LV-Pump interaction model was used to simulate the AV dynamics within the LV cavity (see Figure 6 .4) and to identify AV state by analysing the LVAD current motor waveform. Figure 6 .4: AV state in 2D LV-LVAD realistic geometry simulations. The left pa- nel represents cross-sections of the heart and LVAD during isovolumic relaxation when the aortic valve is closed(AC), with blood flow path indicated by the black arrows. The right panel illustrates the AV open (AO) state during LV ejection. Both panels show LV flow for a pump speed set point of set = 100 rad/s). 150 In another set of simulations, the impeller speed set point was increased from 50 rad/s to 150 rad/s in 50 rad/s increments, then increased to 180 rad/s and 200 rad/s in order to cover the full range of aortic valve state transitions (from ventricular ejection (VE) to aortic valve closed (VC)). The instantaneous pump impeller speed (ω) and motor current (I) for all five set points are shown in Figure and Figure , illustrating that the pump motor current undergoes significant changes at each impeller set point. Figure 6.8 shows the relation between the peak motor current and the instantaneous pump speed, revealing an inverse correlation between the two. In addition, the relationship between motor current, AV closure time and closed duration, exhibited good correlation. As shown in Figure, transition from state VE to state VC occurred with increasing pump speed. Where the VE state corresponded to left ventricular ejection during systole, increasing motor speed set point produced an upward shift of the motor current amplitude, resulting in an increased AV closing time, significantly increasing peak motor current by at least 50% compared to the next lowest speed set point. However, during increase of the set point to 200 rad/s, (see Figure), VC state was continuously maintained, with the distance between AV leaflets decreasing to zero at t= 0.45s, after which the AV remained closed (ANO) with no further blood flow to the proximal aorta. Furthermore, at a set point of 50 rad/s, a low motor current of 0.044A was observed during the AV closing phase (t=1.72s) with a close phase duration of 0.20 seconds. This current then increased throughout the AV open state (t=1.72-2.52s), reaching a 151 maximum value of 0.0733A at t=2.29s. At a set point of 180 rad/s, a low motor current of 0.6246A was observed during the AV closing phase (t=1.77s) with a close phase duration of 0.62 seconds, an increase of 0.42 seconds compared to 50 rad/s. The motor current then increased throughout the AV open state (t=2.05-2.43s), reaching a maximum value of 0.8056A at t=2.29s. Furthermore, at speed set points of 100 and 150 rad/s, low motor currents of 0.150 A and 0.391 respectively were observed during the AV closing phase (t=1.76s) and (t=1.78s), with closed phase durations of 0.38 seconds and 0.52 respectively. The motor current also increased throughout the AV opened state (t=1.90-2.52s, and (t=1.97-2.45s respectively), with delay in open time of 0.14 seconds compared to 150 rad/s. The current reached maximum values of 0.232A and 0.526A respectively at t=1.31s and t=1.28s. However, AV closure durations for set points of 100, 150 and 180 rad/s were increa- sed by 0.18 s, 0.32 s and 0.42 s compared to the 50 rad/s set point, respectively. Ho- wever, the AV fully-open instant at 100 rad/s and 150 rad/s occurred at approximate- ly the same time (t=2.21s), whereas this time occurred earlier for 50 rad/s and 180 r- ad/s at t=2.12s and t=2.17s, respectively. In addition, during AV opening, there were small oscillations in the current waveform at a set point of 180 rad/s, probably due to mechanical flutter of the valve leaflets. As Figure 6.6 illustrates, aortic valve regurgitant flow during high pump speed (i.e., 200 rad/s), occurs during diastole, LV pressure is low with a concomitant reduction in LVAD flow as illustrated in Figure 6.7. It was observed that the pump average 152 flow (Qp=1.4 L/min) at pump speed of 200 rad/s was decreased by 0.6 L/min compared with pump speed of 100 rad/s which was 2L/min Figure , due to the increase in pump speed and the need for a higher LVAD speed to compensate for the back flow of the blood from the aorta, as reported by Rasalingam et al. [171]. These investigators claimed it was caused by aortic root pressure being higher than instantaneous LV pressure as a result of LVAD unloading. In addition, AR, or the inability of the aortic valve to close 153 Figure 6 .5: Simulated electric current and pump impeller speeds at four motor speed set points (50, 100, 150 and 180 rad/s), where ' ■ ' and ' ● ' indicate AV opening and closing times, respectively. Periods 'O' and 'C' represent the phases during which the AV is open, referred to as ventricular ejection (VE), and valve closed (VC), respectively. Max PCT and Min PCT are the maximum and minimum values of motor current during AV closure, respectively. The motor current waveform at Max PCT further decreases once the aortic valve is closed at the closing notch (CN). (a) set = 50 rad/s (b) set = 100 rad/s) (c)set = 150 rad/s) (d)set = 180 rad/s. 154 Figure 6 .6 : Simulated electric current (black line) and distance between aortic valve (AV) leafl-ets (re d line) at a motor speed set points of 200 rad/s, where '●' and '■' indicate AV opening an-d closing ti mes, respectively. The duration of the closing phase was increased from lower set p-oint values, with t he VC state beingcontinuously maintained (ANO). AR denotes the aortic val-ve regurgitant period thr ough the cardiac cycle. Figure 6 .7 : Average inflow rate Qin (dashed line) applied at the inlet (source) boundaries, and the inlet pump flow rate Qp (solid line). 155 completely during diastole, results in the backflow of blood from the aorta into the left ventricle to the LVAD. However, during our simulation, the AR was developed by the result of the high pump speed (i.e. 200 rad/s), low blood flow from the left atrium, the left ventricle is in diastolic phase, and aortic pressure (AoP) is larger than LVP at all times. In addition, the aortic valve was not open during the entire cardiac cycle (ANO), since all the blood flowing thru pump cannula. Therefore, the CO (total flow of mean 1.3 L/min) is equal to the pump flow (see Figure ), meaning that the CO is totally provided by the LVAD at high pump speed. In addition, during AR (t=0.51-0.97s), it was observed that the pump average flow (Qp=1.4 L/min) was increased by 0.4 L/min due to aortic valve regurgitation and the need for a higher LVAD speed to compensate for the back flow of the blood from the aorta, which may lead to premature device failure. In addition, the relationship between motor current amplitude and LV pressure versus motor pump speed (rad/s) exhibited an excellent correlation: as shown in Figure , peak motor current increased with decreases in LVP with the increase in the pump speed. Figure 6.9 shows the relation between the AV closure ti- me and motor pump speed. The closure time increased with the increase in the motor pump speed, indicating that AV closure time reaches a point when stays closed during all cardiac cycle. 156 Figure 6 .8 : LVP–peak (red dotted line) and LVAD motor current (black solid line) against pump speed in LV model. Figure 6 .9 : Closure time against motor speed in LV model 157 Figure and Figure illustrate simulation results of blood flow AV movement during LVAD support, where the closure time is approximately the same at both (50 and 100 rad/s) speed set points. However, the opening time is delayed by 0.18 s at the motor speed set point of 100 rad/s. At 150 and 180 rad/s speed set points, the opening time is approximately the same for both, however the closure time was delayed by 20 ms at the set point of 180 rad/s. 158 Figure 6 .10 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. For each panel, the three snapshots illustrate AV closing phase (left), opening phase (middle) and fully-open (right). (a) set = 50 rad/s. (b) set = 100 rad/s. 159 Figure 6 .11 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. For each panel, the three snapshots illustrate AV closing phase (left), opening phase (middle), and fully-open phase (right). (c) set = 150 rad/s. (d) set = 180 rad/s. 160 6.3.3 Discussion The simplified and realistic LV models described in this chapter were developed with the aim of providing insights into the dynamics of LV-LVAD interaction as well as detecting AV status. The simulation data in Figure show a progressive increase of AV closure duration with increasing pump speed set point and pump electrical current amplitude. During the ventricular ejection state, the LV begins to pressurize, forcing the AV leaflets to open, blood to flow through the aortic valve, and pressure differential head across the pump to decrease. The latter leads to an increase in pump flow and motor current, with a decrease in pump speed that continues throughout systole. However, in the AV closed state, the LV pressure falls, causing AV flow to decrease and differential pressure across the pump (head) to rise. The latter results in a decline in pump flow motor electrical current during diastole. In addition, Figure and Figure illustrate that during both systole and diastole, there was continual blood flow through the pump (see Figure ), and this flow begins to increase as a result of LV contraction at t=1.65s, even before the AV opens at t=1.90s.. The state in which the AV remains closed throughout the entire cardiac cycle, with aortic flow (Qav) being zero and pump flow Qp equal to the CO, indicates that the left ventricular pressure (LVP) is less than the aortic pressure (AoP) and is not enough to open the AV. In our simulations, this state arises when the pump speed is too high, substantially lowering the LV pressure. In 161 terms of simulations of the mechanics of the left ventricle, the real heart exhibits more complex motion and can be modelled with more realistic active contractile properties, such as a time-varying elastance. This may affect the onset of AV open-close times. However, such detailed modelling was beyond the scope of this thesis work. Furthermore, the model may be extended in future to incorporate a range of heart failure conditions including changes in cardiac contractility. Figure 6 .12 :Average pump inflow rate (Qp) at a motor speed set points of 100 rad/s, where '●' and ' ■' indicate AV opening and closing times, respectively. 6.4 Limitations Modelling LV geometry in a congestive heart failure patient remains a challenge, due to the limited available in vivo data of the remodelled myocardium during LVAD support. One option is to compare simulation results with existing studies. The LV has a complex geometry with a smooth endocardial surface and anisotropic 162 material properties. However, our results were obtained using an idealized geometry with smooth surface and isotropic material properties, which may alter the flow patterns in the myocardium. However, in this study, our LV model was used only to supply fluid velocity loading conditions for the AV. Since the behaviour of the ventricle was not the focus of the study, it was not necessary to represent its structural properties accurately. A final limitation of the present model is that the measured LV pressure during ventricular ejection (Figure 6.3) was much lower (i.e. maximum LVP during AV open was 40 mmHg at pump speed 150 rad/s) compared to previous modelling data (i.e. 120 mmHg during AV open) [141]. However, we found that LVP is highly variable from one study to the other, probably due to the simplified LV geometry used, and the simplified ventricular contractility settings during systole. 6.5 Conclusion In this chapter, simulation results were presented from simplified 2D FSI models of the LV in the presence of LVAD, based on the ALE method. The models were formulated with the aim of providing insights into the dynamics of heart-pump interaction, and to simulate LVAD motor current waveform in relation to the AV state. The results confirmed that when the AV opens, there is a higher motor current compared to when the valve is closed. However, our computational model of motor 163 current variations during AV movement, particularly during its closing phase, and under pulsatile flow conditions, will provide significant insights into LV function during LVAD support, particularly as the model is further developed to incorporate physiological heart failure conditions. Moreover, our simulations offer the potential for improving current LVAD control systems to ensure patient safety and comfort, reducing the incidence of AV pathologies during heart pump support, and in helping reduce the risk of aortic insufficiency. 164 7 Chapter 7 – Effect of Parameter Variations on Aortic Valve State under Rotary Blood Pump Assistance 7.1 Introduction The use of a left ventricular assist device (LVAD) as mechanical assistance for heart failure disease is of great importance, due to its ability to stabilize heart function until a suitable donor heart becomes available, or to allow adequate mechanical assistance for the native heart to heal itself. Due to advances in LVAD development, as well as the shortage of appropriate heart donors, LVADs are often used as a bridge to transplant, or more frequently nowadays, a destination therapy [6, 172]. However, since patients undergo different activity levels from sleep to exercise, underpumping or overpumping can often occur, leading to undesired consequences in LV haemodynamics, as well as abnormalities in AV state [168]. Numerical models [92, 173-175], mock loop experimental studies [176-178], animal studies [156, 179] and human studies [180, 181] have been used to examine the effect of LVADs on the cardiovascular system (CVS). On the contrary, FSI models are able to simulate the response of the CVS in the presence of left ventricular support devices, and can provide additional insight into the dynamics of the assisted circulation under different operating conditions. Numerical models also offer a stage for refining existing techniques used to control the speed of the implantable rotary pump. Consequently, monitoring of AV 165 opening and closure during heart pump support is crucial in preventing AV abnormalities and remodelling caused by anomalous recirculation. Previous LVAD experiments have concluded that the ideal control set-point is where left ventricular ejection is occurring and there is a net positive flow through both the aortic valve and the pump [182-184]. However, implanted LVADs alter the haemodynamics of the heart, leading to the necessity of determining the optimal speed set point to satisfy the varying physiological needs of the patient, and to ensure maximum end-organ perfusion [185]. Consequently, the identification of AV state during LVAD support is necessary to prevent complications reported in previous studies such as recirculation and stasis inside the LV cavity [186, 187], as well as aortic valve fusion [7]. Clinically, AV state is measured using echocardiography, whilst aortic flow is assessed using pulsed Doppler Ultrasound after LVAD implantation. Identifying heart state using non-invasive variables has concentrated on LV suction state detection and non-suction state [183, 188, 189], whereas few studies have focussed on detecting AV state during LVAD support using pump motor electrical current [136, 190-192]. Identifying AV state according to the LVAD motor current signal can provide a non- invasive input for a pump speed controller, in order to prevent highly negative pressures developing in the left ventricle when there is insufficient blood in the ventricle to sustain normal left ventricular ejection and the AV remains closed, as reported by Karantonis et al. [183]. In this instance, there is no flow through the AV and the possibility of blood stasis distal to the AV could lead to significant complications from thrombus formation. This 166 chapter will describe results from two LV pump geometry models, in order to evaluate the immediate response of the AV to different physiological states over a wide range of pump speed conditions. Several studies [144, 145, 193, 194] have investigated non-invasive pump motor feedback signals (current or speed), to be used as useful indicators of LVAD pumping state for either overor under-pumping conditions. For example, Yuhki et al. [193] considered a waveform deformation index based on a spectral analysis of the speed signal; Oshikawa et al. [194] and Endo et al. [136] studied the pump motor current amplitude; and Voigt et al. [195] used the differentiated current waveform as an LV wall suction indicator. Most groups failed to show the impact of physiological conditions and other cardiovascular characteristics on the interaction of the ventricular assist device with both the LV and the aortic valve. The results of this chapter are divided into two parts. In the first part, results from a simplified 2D FSI LV finite element model with implantable rotary pump are presented to predict AV state during LVAD outflow. In the second section, a realistic 2D left ventricle geometry is used. These models will be useful tools in the development of a pump speed controller for optimal management of pump outflow. The main results of this chapter were published in Alonazi et al. [34]. 167 7.2 Simplified LV-pump model 7.2.1 Results and Discussion In this section, we present results from a simplified 2D model of LVAD support, investigating for the first time, the effect of variations in cardiovascular parameters on AV open and closed states during LVAD assist. As demonstrated in Figure 7.1, model behaviour could effectively be divided into two phases: aortic valve closed and open. In this simulation, the impeller speed set points were 100 and 150 rad/s, covering the range of aortic valve state transitions from ventricular ejection (VE) to aortic valve closure (VC). Figure 7.1 shows the instantaneous pump motor current (I) for both set points for two abnormal physiological conditions, H2 and H3, compared to the H1 (healthy) condition, illustrating that the pump motor current undergoes significant changes for each condition. Furthermore, there was a good correlation between motor current, AV closure time and AV closure duration. 168 Figure 7 .1: Simulated electric current and pump impeller speed at two motor speed set points of 100 rad/s (top) and150 rad/s (bottom), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively, the periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the periods during which the AV is open and closed, respectively at conditions H1, H2 and H3 respectively. MAX C1PT, MAX C2PT and MAX C3PT, and MIN C1PT, MIN C2PT and MIN C3PT are the maximum and minimum values of motor current during AV closure. 169 As shown in Figure 7.1 reduced Vtotal (H2 condition) produced a downward shift of the current amplitude, resulting in an increased AV closing time, significantly decreasing peak motor current by 50% compared to the healthy H1 condition under both speed set points. Furthermore, at both set points of 100 and 150 rad/s, high motor currents of 0.31A and 0.67A respectively were observed for H1 (t=2.00s), decreasing rapidly throughout the AV c losed state (t=2.21-2.70s and t=2.20-2.75s respectively), reaching a minimum value of 0.09 5A at t=2.60s and 0.295A at t=2.53s respectively. Figure 7.2 shows simulation results of AV movement during LVAD support, where the op- ening time at 100 rad/s was delayed by 0.12 sec for H2 compared to H3. However, opening time remained approximately the same for both H2 and H3 at 150 rad/sec. However, the closure time at speed points of 100 and 150 rad/s was delayed by 0.1 sec and 10 ms respectively between the two conditions, with the closure period under H3 and H2 being between 0.48 sec and 0.70 sec, and 0.67 sec and 0.66 sec, respectively, compared to the H1 values of 0.45 sec at 100 rad/s and 0.57 sec at 150 rad/s. 170 Figure 7 .2: Simulated distance between AV leaflets at two motor speed set points of 100 rad/s (top) and150 rad/s (bottom), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. The periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent phases during which the AV is open and closed, respectively, for conditions H1, H2 and H3 respectively. 171 However, the closure time was increased by 10 ms and delayed by 30 ms compared to H3, and was delayed by 0.1 sec and increased by 40 ms compared to H2. In addition, during AV opening, there were small oscillations in the current waveform, probably due to mechanical flutter of the valve leaflets. Figure 7.2 shows the AV dynamics at different phas es in the cardiac cycle for the three physiological conditions (H1, H2 and H3), starting at t- he onset of AV opening and ending at AV closure. AV state during these physiological co- nditions was characterized by the measured distance between AV leaflets as shown in Figure 7.2. Under each condition, the impeller speed was fixed at two speed set points of 100 and 150 rad/s. The figure shows the relation between the peak opening distance and the total duration of opening and closing AV times, revealing the correlation between the pump speed and distance between leaflets under each physiological condition. Results of blood flow and AV movement are shown in Figure 7 .3 for a speed set point of 100 rad/s. 172 Figure 7 .3: Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle for = 100 rad/s. (a) Aortic valve closing and opening phases under condition H1. (b) Aortic valve closing and opening phases under condition H2. (c) Aortic valve closing and opening phases under condition H3. 173 7.3 Realistic LV-Pump model 7.3.1 Introduction In this section, we examined the response of the aortic valve (i.e. open and closed) during a working range of pump speeds and changes in physiological condition. The realistic 2D LV-Pump interaction model described in Chapter 6 was used. The main finding of this set of simulations is the influence of rotary pump output on AV state during three physiological conditions; healthy and two different physiological conditions simulated by altering cardiovascular parameters: i. by varying systemic vascular resistance (SVR) ( or afterload) (condition H3) ii. by changing the total blood volume Vtotal (or preload) (condition H2), by varying the rate of inflow to the model through the source inlet. The pump set speed was increased from a speed of 50 to 150 rad/s 7.3.2 Results: To investigate the effect of motor pump current on AV open and closed states, the response of the AV under LVAD support was examined under three physiological heart conditions using the realistic 2D LV-pump interaction model. As shown in 174 Figure 7 .4: Simulated motor electric current at three motor speed set points (50, 100, and150 rad/s), where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. The periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the phases during which the AV is open and closed respectively, at conditions H1, H2 and H3 respectively. 175 Figure 7.4, model behaviour could effectively be divided into two phases: aortic valve closed and open. In this simulation, the impeller speed set points were set to (50, 100 and 150 rad/s), covering the full range of aortic valve state transitions from ventricular ejection (VE) to aortic valve closed (VC). To obtain the individual effect on AV state of two abnormal physiological conditions, H2 and H3, compared to the H1 (healthy) state, pump impeller motor current (I) was simulated for all speed set points for all three physiological conditions, H1, H2 and H3, revealing that the pump motor current undergoes significant change under each condition. In addition, the relationship between motor current, AV closure time and AV closure duration exhibited good correlation. As shown in Figure 7.5, reduced Vtotal formed a downward shift of the current amplitude, resulting in a reduction in AV closing time by 80 ms, significantly decreasing peak motor current (MAX O2 PT) by approximately 50% compared to t- he healthy condition (H1) maximum current (MAX O1 PT) at 50 rad/s speed set po- ints as shown in Figure 7 .5(a). On the contrary, at speed set points of 100 and 150 rad/s, AV closing times were increased by 130 and 160 ms, respectively, compared to H1, whereas, at condition H3, the peak motor current (MAX O3 PT ) was increa- sed slightly, but was almost the same as to the healthy condition under both speed s- et points. Moreover, at speed set points of 100 and 150 rad/s, high motor currents ( MAX O1 PT) of 0.23A and 0.52A, respectively, were observed for H1 (t=2.28s) du- ring AV open state, decreasing rapidly throughout the AV closed state (t=2.52-2.90 s and t=2.45-2.98s respectively), reaching a minimum value (MIN C1 PT) of 0.151 A at t= 2.77s and 0.391A at t=2.78s, respectively, (see Figure 7 .5 (b and c)). 176 Figure 7 .5: (a), (b) and (c). Simulated motor current waveform during AV closure/open phases, where ' ● ' and ' ■ ' indicate AV opening and closing times, respectively. Current levels MAX C1PT, MAX C2PT and MAX C3PT, and MIN C1PT, MIN C2PT and MIN C3PT denote the maximum and minimum values of motor current during AV closure, whilst MAX O1PT, MAX O2PT and MAX O3PT are the maximum current values during AV open phases. From top to bottom, the panels show pump speed set points of (a) set = 50 rad/s, (b) set 100 rad/s and (c) set = 150 rad/s). 177 Figure 7.6 shows simulation results of AV movement during LVAD support, where the opening time at 50 rad/s was delayed by 10 ms more for H2 than for H3, but r- emaining approximately the same for both H2 and H3 at 100 rad/sec. The closure time at speed points of 50 and 100 rad/s was shifted by 0.1 s and 40 ms respectively between the two conditions. AV closure periods under conditions H2 and H3 were between 0.13 sec and 0. 11 sec, and 0.27 sec and 0.31 sec, respectively, compared to the H1 values of 0.20 sec at 50 rad/s and 0.34 sec at 100 rad/s. Figure 7 .6: Simulated open and closed states of AV leaflets at two motor speed set points of (a) 50 rad/s and (b) 100 rad/s). The waveforms show the distance between AV leaflets, where the distance in the open state the distance was measured with a positive value in millimeters. This open state is referred to as ventricular ejection (VE), whilst the valve closed (VC) state is when the distance was zero. Periods 'O1', 'O2', 'O3' and 'C1', 'C2' and 'C3' represent the phases during which the AV is open and closed, respectively at conditions H1, H2 and H3 respectively. 178 The closure time was delayed by 70 ms and decreased by 30 ms compared to H3, and was decreased by 70 ms and shifted by 90 ms compared to H2. Figure shows the blood velocity and AV dynamics at different phases in the cardiac cycle for all three physiological conditions, starting at the onset of AV opening and ending at AV closure. Generally, we observe that with all conditions, AV opening time was reduced and the closure time was increased with the increase in pump speed from 50 rad/s to 100 rad/s. 179 Figure 7 .7 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle for = 150 rad/s. (a) Aortic valve closing and opening phases under condition H1. (b) Aortic valve closing and opening phases under condition H2. (c) Aortic valve closing and opening phases under condition H3. 180 7.3.3 Discussion Endo et al. [136] and Oshikawa et al. [194] investigated the variation in pump motor current amplitude to estimate motor pulsatility, without using specific sensors, and ignoring the AV response to CVS parameter variations. However, in this chapter, the impact on AV dynamics using an LV-pump interaction model with varying degre- es of LVAD assistance under various heart failure conditions was investigated, demonstrating that electrical motor current amplitude alters with variation of cardiova- scular parameters due to changes in AV state. The model was able to simulateAV state during three physiological conditions across the full range of pump speeds (i.e. 50, 100 and 150 rad/s). Figure 7.6 shows various AV dynamics states in one cardiac cycle. The AV state during different physiological situations was characterized by the measured distance between aortic valve leaflets. The relationship between the peak distance and duration of opening and closure AV times revealed the correlation between the pump speed and extent of AV opening under each physiological condition. During the healthy H1 condition, the leaflets were able to reach a full opening diameter of 25.3 mm due to myocardial contraction, causing the left ventricular pressure to increase (LVP) and expulsion of blood from the ventricle. This result was perfectly matched by in vivo measurements of AV opening diameter in the human heart of 24.4 mm, as measured by Doppler Ultrasound [196]. 181 In terms of simulating the mechanics of the LV, the actual heart is known to exhibit more complex motions and can be modelled with more realistic active contractile properties, such as a time-varying elastance. This may affect AV open-close times. However, such detailed modelling was beyond the scope of this study. This simple method of classifying was based not only on the time of the systole as with other studies [191], but also on the diastolic portion of the pump current signal, and allowed a clear distinction between open and closed AV states in real time. Such information will lead to a more precise evaluation of AV function, helping to reduce the risk of aortic insufficiency. Furthermore, the model can be extended to incorporate a range of heart failure conditions including changes in cardiac contractility. Currently, the only way to confirm myocardial recovery and ensure the opening of the aortic valve during LVAD support is to perform serial echocardiograms and adjust the pump speed accordingly [197]. However, our AV dynamics models could be used in developing a controller with the objective of maintaining the pump speed set point within safe pump operating ranges for heart recovery. Such controllers can estimate if the AV opens and closes normally at a given pump speed. As reported by Maybaum et al. [40], partial loading of the LV during LVAD support (i.e. pump assistance ratio < 100% where there is net positive flow through both the aortic valve and the LVAD), is associated with improved myocardial function. 182 7.4 Model Limitations Due to the lack of existing aortic valve and left ventricular assist device interaction simulation and experimental studies, more data is needed to understand the native heart responses to the changes in physiological parameters associated with varying levels of HF. In the present model, we have not taken into account adding more physiological parameters associated with varying levels of contractility, such as exercise and postural changes, due to their complexity and to understand the native heart responses to changes in these parameters associated with HF. However, careful attention would have to be taken while modelling the response in a chronic heart failure patient, since it may be significantly different from that of a healthy subject [198]. 7.5 Conclusions In this chapter, we have investigated the effect of alterations in model parameter values, namely SVR and Vtotal, on AV state under rotary pump assistance, using both a simplified and realistic 2D LV-pump interaction model. In particular, the models were able to simulate the response of the AV to changes in LV preload and afterload in the presence of LVAD support for the proper management of pump speed settings. 183 8 Chapter 8 - Simulation of Motor Current Waveforms as an Index for Aortic Valve Condition during Ventricular Support Existing commercially-used left ventricular assist devices (LVADs) make no attempt to automatically detect the aortic valve condition in their control methods to optimize ventricular assistance. However, to develop such a control strategy, an important design goal for iRBPs is the ability to reliably and accurately detect AV states that can cause harmful effects on AV structure and function due to aortic valve backflow (regurgitation) as a result of aortic root pressure being always higher than LVP as a consequence of LVAD over-pumping [171], and the stress applied on AV leaflets due to pump back flow (regurgitation) as a result of under-pumping [199]. In addition, varying LVAD pump speed can control the state of the aortic valve. By increasing impeller speed, it is possible to transition from the normal physiological state of AV opening during ventricular ejection to a state where the aortic valve remains permanently closed throughout the cardiac cycle. The problem of AV state detection has attracted substantial research interest [182, 189, 190, 197, 200- 202]. However, none of these studies have included aLV-pump FSI model to investig- ate the impact on the AV during left ventricular assist. 184 FSI and CFD have emerged as reliable tools that can provide mechanical and dynamical field information as well as other variables essential for the assessment of cardiac function under assist device support. The FSI method is utilised in this study for examining the AV response during LVAD support. However, limited studies have focused on identifying the aortic valve state, despite its importance in the improvement of current LVAD control algorithms, which may offer better response times and more normal cardiac output aiming for myocardial recovery to ensure patient safety and comfort, as well as reduce the incidence of AV pathologies during heart pump support. In this chapter, we have investigated the correlation between the AV performance and LVAD motor current as well as speed set points; simulating aortic valve blood flow, pressure, pump flow and LV mechanics using a 2D LV-pump interaction model. The model does not include physical AV leaflets, but these were replaced by a valve flow equation and pressure gradient between chambers to mimic the AV function, for more efficient computation of left ventricular pressure (LVP) and AV pressure (AoP) in the model. 8.1 Introduction Important improvements have been achieved in recent times in FSI simulations of cardiac valves. Many problems in cardiac mechanics can be modelled as the dynamic interaction of a fluid (the blood) and an elastic structure (the valves of the heart). Over 185 the past years, medical societies have mostly accepted that valve dysfunction in LVAD patients may be related to structural changes in the tissue induced by altered biomechanics and excessive stress, and a reliable assessment of full assist state requires an estimation of the aortic valve mechanism during LVAD support [203, 204]. In this chapter, we evaluate AV state using the LVAD motor current and aortic valve flow waveforms as an index, simulated from a 2D LV computational model without AV leaflets under heart pump support. The objective was to investigate how motor electrical current waveform changes with AV state and with the increase in pump speed. Also, identifying AV state according to motor current signals may provide an additional input to a pump speed controller to prevent highly negative pressures in the left ventricle, leading to wall suction, as well as ensuring the AV opens and closes over the cardiac cycle. The hypothesis here is that by using the FSI 2D LV-pump model and only the non- invasive measures of instantaneous pump electrical current and impeller speed, it is possible to automatically detect AV states including: AV open state during Ventricular ejection (VE) AV closed state during diastole (AC) AV not open state (ANO), remaining closed throughout the entire cardiac cycle. 186 AV regurgitation state (AR), which frequently occurrs in heart failure patients assisted by LVADs, caused by high pump speed. Results show that when the AV opens, there is a higher motor current compared to when the valve is closed. Also with the increase in pump speed, the AV state shifts from the VE into the VC state. With high pump speeds, the AV shifts into the ANO state. In addition, we observed a negative flow jet through the cardiac cycle which oc- cured at lower relative pump speed (i.e; 50 rad/s) causing a pump regurgitant state. 8.2 Simplified LV-pump model 8.2.1 Results: Figure 8.1 shows the waveforms obtained from a simplified 2D LV-pump interaction model simulation. An initial inspection of the invasive variables of LVP, AoP, aortic valve flow (Qav), pump flow (Qp) as well as the non-invasive variables of motor electrical current (I) and pump impeller speed, indicated the presence of AV physiological open/closed states under LVAD support. Three pump speed set points are illustrated. The AV was open during systole, where blood flowed from both the left ventricle and the pump, as seen from the positive Qp and Qav throughout the cardiac cycle. VE was characterized by LV ejection during systole, with greater LVP compared to Aop (LVP > AoP), and positive aortic valve flow (Qav > 0). 187 In addition, during systole, the contracted LV model produced a low pump speed and high motor current, causing the pressure differential across the pump to decrease and pump flow to increase. Torque is proportional to input power for a constant coil drive voltage. This means that as torque fluctuates, so does motor current and input power. As flow increases, it causes the impeller speed to fall and power to rise. This result is similar to that obtained in previous animal [183] and computational studies [149]. In addition, Figure 8.1 shows that the transition between AV states was induced by changes in LV wall contraction. During the end of the LV pressurization (diastole), the aortic valve opened and the LVP sharply declined. Aortic pressure gradually decreased, causing a rise in differential pressure across the pump (head) and Qp to fall away much later than the Qav (approximately 0.27 s) during diastole, with less torque on the impell- er leading to an increase in speed. Thus, as shown in Figure 8.1, the current waveform in both diastole and systole shows a sinusoidal profile, equally uniform, with adjacent cardiac cycles symmetrical between systole and diastole, and inversely related to the motor speed waveform (i.e. as current increases, speed decreases). In addition, during AV opening at a pump speed of 150 rad/s, there was an increment in the current amplitude and small oscill- ations in the current waveform, probably due to turbulent blood flow patterns. On the other hand, the motor current waveform showed at low pump speed set point (i.e. 50 rad/s) along with a 50% decrement in amplitude compared to 150 rad/s. 188 Figure 8 .1 : Simulated aortic valve flow (Qav), pump flow (Qp), left ventricular pressure (Plv) and aortic pressure (Pao), motor current and pump speed waveforms obtained from the combined simplified LV-Pump model. Three pump speed set points (50, 100, and 150 rad/s) are shown. 189 Figure shows simulation results using the simplified LV-pump model without AV leaflets during LVAD support. The motor speed set points start from 50 rad/s and increase by 50 rad/s to 150 rad/s, where the opening time at lower motor speed (i.e. 50 rad/s) occurs much earlier than at high motor speed set point (i.e. 150 rad/s) approximately by 0.12s, and 70ms less than that at 100 rad/s. However, at a low speed set point of 50 rad/s, the closure time was delayed by approximately 0.1s than at a motor speed set point of 150 rad/s and by 40ms compared to 100 rad/s. 8.2.2 Discussion This set of simulations aimed to develop a simplified LV-pump model capable of predicting the time course of simulated haemodynamic variables in a left ventricular model. Furthermore, the model of the aortic valve and the rotary blood pump serves as an important platform for the next stage of modelling realistic geometries of a LV- pump model to investigate the aortic valve response under LVAD support. It was observed that during high speeds (i.e.150 rad/s), LV contractions eventually dec- reased and the model became akin to LV suction, as shown in Figure 8.2. However, fl- uctuations in pump flow continued to occur due to turbulence as a result of the high motor speed (i.e. 150 rad/s). 190 Figure 8 .2 : Snapshots of simulated LV blood velocity magnitude during LVAD support at various phases during the cardiac cycle. (a) = 50 rad/s. (b) = 100 rad/s. (c) = 150 rad/s. 191 8.3 Realistic LV-Pump model Results In this section, we presente results from the realistic 2D LV-pump FSI model without AV leaflets. Due to the complexity of the real system, LV geometry and associated structures have been idealised. The model did not include physical AV leaflets: these were replaced by a valve flow equation to mimic the AV function for more computationally-efficient simulations of left ventricular pressure (LVP) and AV pressure (AoP). 8.3.1 Identification of AV state In order to evaluate whether pump speed and motor current are potential indicators of aortic valve state transition, it was necessary to independently verify the physiological state of the AV in the simulations. Since this model did not include the AV leaflets, A- V state was identified from model pressure and flow data, as well as pump intrinsic variables, as indicated in Table 8 -1. Model behaviour can effectively be broken up into two phases, systole and diastole. An initial inspection of the invasive variables of aortic valve flow and the difference between left ventricular pressure (LVP) and Aortic pressure (AoP) indicated the prese- nce of open and closed aortic valve states (see Figure 8.4, top panel). 192 By using only the non-invasive measures of instantaneous pump impeller current to evaluate AV dynamics during LVAD support, it may be possible to detect a range of AV states, as described in previous chapters. In this set of simulations, attempts were made to identify the AV states using the criteria of Table 8.1. Table 8 -1. A summary of physiological and pump basic parameters identifying the aortic valve conditions. Physiological Pump Basic Variables Variables AV State Motor LVP Pump speed Qav current AC decreasing VE increasing >AoP decreasing >0 8.3.2 LV wall motion Figure 8 .3 : Left ventricular wall simulated cavity dimension during contractile motion. 193 Measurements of end-systolic cavity dimension and end-diastolic cavity dimension were compared with echocardiographs data from human experiments [205]. However, since our model apex structure is fixed, and the LV walls were the only moving boundaries, the model’s LV wall contraction was less when compared to a real heart failure subject. Therefore, we have analysed only short axis displacement by measuring the horizontal displacement of the LV cavity as an alternative. In our study, the cavity dimension was measured as the maximum short axis displacement between the septal endocardium and the posterior left ventricular wall, with dimension results shown in Figure . Due to the impact of the LVAD on CVS dynamics, the maximum short axis displacement of the model was 3.5 mm less than cavity displacement dimension magnitude reported in a prior human echocardiography study for heart failure subjects, which was 20 mm [253]. 8.3.3 LV-Pump model simulations To examine the change in model haemodynamic and pump variables with increasing pump speed set point under fixed LV contractility, values of key variables from our LV-pump interaction model study were plotted in Figure 8.4 and Figure 8.6. An initial inspection was made of the invasive observations of LVP, AoP, Qav , as well as the non-invasive variables of motor electrical current (I) and pump impeller speed. During pump assist, most variables showed a change in the amplitude with respect to pump speed. Qav was lower with increased pump speed (see Figure ). In addition, at a high pump speed set point (i.e. 200 rad/s), Qp was higher due to lower differential pressure 194 across the pump (head), producing a higher pump assistance as reported by Maybaum et al. [40] (i.e. pump assistance ratio = 100%) and transition into ANO state, as shown in Figure (d). Figure 8.4 illustrates the waveforms obtained from a realistic 2D LV-pump interaction model simulation. Four pump speed set points were applied to the model from 50 rad/s to 200 rad/s in increments of 50 rad/s. The figure shows pressure patterns (LVP and AoP) waveforms during speed changes in this 2D LV-pump model simulation. 195 Figure 8 .4 : Top panels: left ventricular pressure (black line) and aortic valve pressure (dotted red line) at variouspump speeds. Lower panels: Corresponding pump speed waveforms from the LV-Pump model. Four pump speed set points (50, 100, 150 and 200 rad/s) are shown. 196 The simulation data showed a decrease in pulsatility in aortic pressure with increasing pump speed until the point where the AV remained permanently closed ( Figure , 200 rad/s). This was because LVAD support directly correlated with unloading of the LV while maintaining aortic blood pressure, so this loss of pulsatility is as expected during LVAD support. This finding is similar to the experimental studies of Goldstein et al. [179] and Choi et al. [206]. Figure illustrates the motor current and aortic flow waveforms obtained from four different set points of 50, 100, 150 and 200 rad/s, illustrating that the pump motor current undergoes significant change for each impeller speed set point. The result indicates the relationship between the peak impeller speed and Aop in our 2D LV-pump interaction model, revealing an excellent correlation with the aortic pressure. In addition, due to absence of a mitral valve in the 2D realistic LV model, we have found that the LVP and Aop are lower than observed in a previous animal study by Karantonis et al. [183] and a computational study by Lim et al. [149]. However, our low left ventricular pressure result was similar to that obtained from a numerical model study by Shi et al. [135] for investigating human CVDS response to LVAD, as well as previous animal [207] and mock-loop [208] experiments. In addition, Figure 8.4 sho- ws time series plots calculated from the pump speed. The plots were obtained from our pump model at a certain value of the pump speed set points, providing another useful representation of the pump flow state characteristics. 197 Figure 8 .5 : Simulated aortic valve flow (Qav), (black line) and pump flow (Qp) (red line) at a pump set point of 100 rad/s. Contraction in the model results in distinct AV open/close phases. For example, the ideal AV state appeared to be that in which the aortic valve opened during systole, where blood flowed from both the left ventricle and the pump, as seen from the positive Qp and Qav flows during systole (see Figure 8.5). Contraction of the ventricle caused th- e differential pressure across the pump to decrease, and the force or torque on the impeller to rise. As pump flow increases, it causes impeller speed to fall and electrical power to rise. In contrast, during diastole, impeller speed increased, causing motor current to decrease. This result is similar to that reported in previous animal [183] and computational studies [149]. From the pump motor current and aortic valve flow signals in Figure (a, b and c), model behaviour could effectively be divided into two states: aortic valve closed and open, similar to the previous simulation results. Furthermore, the relationship between motor current, AV closure time and closure duration exhibited good correlation. 198 Figure 8 .6 : Simulated pump motor current (black) and reference aortic flow (Qav) (red) at four motor s- peed set points corresponding to (a) 50 rad/s, (b) 100 rad/s, (c) 150 rad/s and (d) 200 rad/s, where '■' and '●' indicate AV opening and closing times, respectively. Qav > 0 represents the periods when the AV is open, referred to as ventricular ejection (VE), and Qav =0 represents the valve closed (VC) state. Note that in panel (d), Qav = 0 at all times, indicating the AV remains closed throughout the cardiac cycle (ANO) for this pump set point speed of 200 rad/s. As shown in Figure , transition from state VE to state VC occurred with increasing pump speed. In addition, increases in motor speed set point produced an upward shift of the motor current amplitude, resulting in an increased AV closing time, and significantly increasing peak motor current by at least 50% compared to the next lowest speed set point. However, at the increase of pump speed set point to 200 rad/s, 199 the aortic valve not opening (ANO) state was continuously maintained, with the AV remaining closed throughout the entire cardiac cycle (see Figure , d), with no blood flow to the proximal aorta. Such an ANO state during the LVAD support has been reported by limited studies [136, 182, 183, 190-192], occurring when AoP is continually higher than LVP with no sufficient pressure in the LV cavity to open the AV (see Figure 8.4, top panels). Furthermore, a low motor current of 0.021A was observed at 50 rad/s in the AV closing phase, with this phase starting at t=1.89s and lasting for a duration of 0.57 seconds. Motor current increased throughout the AV open state (t=1.97-2.40s), reaching a maximum value of 0.052A at t=2.37s. In contrast with 150 rad/s, a low motor current of 0.375A was measured during the AV closing phase (t=1.70s), with a closing duration of 0.78 seconds, representing an increase of 0.21 seconds compared to 50 rad/s. The motor current increased throughout the AV open state (t=2.13-2.34s), reaching a maximum value of 0.4947A at t=2.25s. Furthermore, at a speed set point of 100 rad/s, a low motor current of 0.1143 A was observed during the AV closing phase (t=1.65s), with this phase lasting for a duration of 0.74 seconds. Motor current then increased throughout the AV open state (t=2.11-2. 35s), with an increase in the total open time of 30 ms compared to the 150 rad/s set point. Motor current, then reached a maximum value of 0.196 A at t=2.24s. The total closure durations for pump speed set points of 100 and 150 rad/s were increased by 0.1 and 0.21 s compared to the 50 rad/s set point, respectively. However, as shown in Fig- 200 ure 8.11, the fully-open AV instant at 50 rad/s and 100 rad/s remained approximately the same (t=2.21 and t=2.22s respectively), whereas it was increased at 150 rad/s to o- ccur at t=2.27s. In addition, we observe in Figure 8.6 a rightward shift of the pump motor current from the aortic flow waveform, particularly, at AV fully-open times, at all four pump speeds, most likely due to the motor winding inductance. In Figure 8.9a, we observed a negative flow jet through the cardiac cycle which occu-r ed at relatively low pump speed (i.e. 50 rad/s) causing a pump regurgitant state at t=1. 97s. This LVAD behaviour during low pump speed has been reported by previous studies including that of Yuhki et al. [193] and Karantonies et al. [183]. This pheno- menon was due to negative pump flow during VE (see Figure ), resulting in AoP being higher than the LVP and the pump differential pressure (head) as shown in Figure . Figure 8 .7 : Simulated aortic valve flow (Qav), (black line) and pump flow (Qp) (red line) at a pump set point of 50 rad/s. 201 Figure 8 .8 : Simulated left ventricular pressure (sold black line) and aortic pressure (dotted black line) and pump differential pressure head (red line) at the low pump speed set point of 50 rad/s. 202 Figure 8 .9 :Snapshot of simulated LV blood velocity magnitude during LVAD support at various phases i.e. AV closure time (left figure in all panels), LV opening phase (two figures at the centre in all panels) and next cycle closure phase (right in all panels) during the cardiac cycle. (a) = 50 rad/s. (b) = 100 rad/s. (c) = 150 rad/s. 203 8.3.4 Discussion Despite extensive studies previously being carried out on AV state detection using intrinsic LVAD parameters [136, 190, 191, 207], and studies to detect pump state [182, 183, 188, 201], only a limited number of studies have been conducted to automatically detect AV state using LV-pump FSI models. In the simulations of his chapter using a 2D LV-pump interaction model without AV leaflets, we concluded that AV dynamics were affected by the pump speed set point. With increasing pump speed, AV closing duration increased and open time was decreased, eventually reaching a condition at high motor speeds where the AV remained closed throughout the entire cardiac cycle. In order to evaluate whether pump current is a potential indicator of aortic valve state transition, it was necessary to independently verify the state of the aortic valve in the simulations. This was achieved through the use of aortic valve flow and pressures. In addition, using the motor electrical current signal, it was feasible to determine AV state as open, closed, or permanently closed. In addition, using the motor electrical current signal, it was feasible to determine AV state as open, closed, or permanently closed. Although beyond the scope of this thesis, there is undoubtedly a need for more experimental studies in data-driven learning strategies for the identification, classification, and recognition of AV state from pump current waveform. Nonetheless, the modelling paradigm presented in this chapter should be viewed as a first stage in the validation of suitable AV state detection techniques for more robust LVAD control. 204 These results agreed with published experimental findings, which showed an increase in aortic pressure with increasing pump speeds [209]. However, the increase in the AoP and LVP in our simulation results was not as high as that reported clinically. This may be due to the fact that we have used a simplified geometry and have not utilized a sufficiently accurate model of LV contractility. The model described in this chapter will prove useful in determining safe pump speed ranges that can detect whether the AV open and closes normally at a given pump speed. In addition, we have been able to identify pattern changes in the LVAD electrical current, which may indicate if the AV is open, closed and continually closed during the cardiac cycle. Hence, the model will provide a useful tool for developing physiological responsive pump control strategies which could promote myocardial recovery, minimizing the risk of LV pathologies in LVAD patients. 8.4 Model Limitations A possible limitation of our AV state detection method was that the measurement of either electrical current or power was based on fixed pump speed set point, ignoring transient changes in set point characteristics. However, is not likely that set point would need abruptly to change in vivo, even during sudden perturbations of the cardiovascular system during pump operation. 205 Another limitation is that it has been reported that haemodynamic responses of the L- VAD may be different under chronic heart failure and failing heart conditions [179]. Therefore, further simulation using our LV model with induced chronic heart failure (i.e. enlarged heart) could be carried out to more accurately represent a wider range of clinical scenarios. 8.5 Conclusions We have presented a simplified two-dimensional FSI model of the ventricle during LVAD support, without explicitly modelling the AV leaflet structures. The model was formulated with the aim of providing computationally efficient simulations that can provide insight into the dynamics of heart-pump interaction, and to simulate LVAD motor current waveform in relation to AV state. The results confirmed that when the AV opens (i.e. aortic outflow), there is a higher motor current compared to when the valve is closed. This motor current result is similar to that obtained by Lim et al. [149]. Our computational model of motor current variations during AV movement, particularly during its closing phase and under pulsatile flow conditions, will provide significant insights into LV function during LVAD support, particularly if the model is further developed to incorporate realistic anatomies. Moreover, our model is an important tool for improving current LVAD control systems to ensure patient safety and comfort, and reduce the incidence of AV pathologies during heart pump support 206 9 Chapter 9 - Conclusions and Recommendations 9.1 Conclusions Despite extensive research in the area of implantable rotary blood flow devices, interaction between the aortic valve and the heart assist device has not yet been completely understood, to improve the treatment of heart failure patients. Computational models of LVAD-AV interaction are economical and fit for investigating AV dynamics under LVAD support. For example, specific pump parameters that difficult to acquire in-vivo, or the impact of certain pump outflow settings can be determined using LV computational models. This thesis was able to address the aims the study in offering insights into AV state detection from LVAD variables such as pump speed and motor current. Chapter 5 presented a simplified two-dimensional LV-pump interaction model of aortic valve motion, based on the the use of the Arbitrary Lagrangian–Eulerian Finite Element Method and the time-dependent Navier-Stokes formulation of an incompressible viscous fluid. The simulated pressure result was altered during AV leaflet movement, particularly during AV closure, providing significant insights into LV function under LVAD assistance. In the second part of Chapter 5 the simplified model of LV-pump was combined with a Windkessel model of the systemic 207 circulation and heart contractility to correctly simulate AV open state during ventricular ejection (VE) and AV closed state (AC) during diastole. Chapter 6 presented an approach for detecting AV states during LVAD support from basic pump variables (i.e. motor current and pump impeller speed), using a simplified 2D LV-pump interaction model. In addition, methods for estimating the average pump differential pressure (head) and flow rate from pump impeller speed, input power and fluid viscosity were presented for modelling the LVAD. This pump model is able to provide valuable insights into how various LVAD speed set points affect AV dynamics, whereas very few studies have examined AV state in simplified computational model with pulsatile LV contraction. In section two of this chapter, 2D realistic LV-LVAD geometries were used to simulate LV dynamics and pump motor current over a range of pump operating points. It also allowed ascertaining the performance of the AV under modified pump operating conditions. Chapter 7 simulated the effect of alterations in CVS parameter values, namely total blood volume (Vtotal) and systemic vascular resistance (SVR), on AV close and open times during LVAD support. In addition, LV-pump parameters were altered to examine the correlation between simulated motor current and distance between AV leaflets over a wide range of pump speed operating points and various CVS conditions, which may be encountered by LVAD patients in their daily activities. It was shown from the motor current simulation results that the AV open and closed durations were altered by these parameter changes. 208 Chapter 8 presented a number of model CVS dynamics features in a simplified LV- pump model after excluding the AV leaflets. This computationally-efficient model provided important insights into AV and LV behaviour under LVAD support. The valve leaflets were substituted with a pressure gradient equation to describe aortic valve flow. This simplified model was shown to be able to predict the time course of simulated haemodynamic variables and pump intrinsic parameters (i.e. motor current and impeller speed), as well as AV state, during LVAD support. 9.2 Suggestions for Future Work While this thesis has addressed many of the important issues associated with AV monitoring and detection in the LVAD-assisted heart, there were numerous areas that still need to be investigated. These include the ability to simulate AV state under a wider range of cardiovascular states and pump operation conditions (i.e. myocardial contractile dysfunction), during exercise and postural changes with more realistic heart geometry. In addition, in order to validate the outcomes of the AV state detection resulting from our model, further in-vivo and in-vitro experiments are required. Furthermore, the ability of the model to detect the level of blood viscosity or HCT from pump intrinsic parameters (i.e. motor current waveform) is an interesting area that needs further analysis. 209 9.2.1 Simulating chronic heart failure during exercise and postural change Thus far, various states of heart failure severity have been defined by different levels of myocardial contractility, changes in venous return blood volume and different vascular resistance values (i.e. Chapter 7). A worthwhile goal for future studies would be to expand the range of HF. One limitation of our model is that myocardial contractility was fixed, and not able to be descreased (i.e. contractile dy sfunction), which is usually determined through a decrease in wall motion. Alterna- tively, methods can be used to achieve this objective (i.e. virtual heart with myocardial fibre direction), and to more accurately simulate chronic heart failure by modelling the LV with patient-specific realistic cardiac morphologies captured from non-invasive imaging modalities such as computed tomography, magnetic re- sonance imaging or echocardiography. 9.2.2 AV state detection validation Evaluating the suitability of the proposed AV detection strategy was performed in our study purely via software simulation. Although the model was validated with limited previously developed in-vivo and in-vitro studies, further investigation and validation of the LV-pump and AV dynamics results using animal experiments under various physiological conditions, including postural changes and HF severity are required. However, this experimental validation was beyond the scope of this thesis. 210 9.2.3 Effect of blood viscosity on AV states As been reported from various in-vitro studies [65, 69], different values of haematocrit (HCT) may affect the estimation of pump flow rate at the same pump power and impeller speed. In addition, it was observed from in-vitro experi- ments by Wakisaka et al. [62] in the goat, that there is a strong correlation betwe- en HCT levels and estimated pump flow. Chapter 6 of this thesis described a model to estimate pump head differential pressure and flow rate from LVAD pump power and impeller speed, which depends on changes in blood viscosity. Fu- ture studies should investigate the effect of blood HCT or viscosity on AV state during LVAD support. 211 Bibliography [1] A. S. Go, et al., "Heart Disease and Stroke Statistics--2014 Update: A Report From the American Heart Association," Circulation, vol. 18, p. 18, 2013. [2] S. Mendis, et al., Global atlas on cardiovascular disease prevention and control: World Health Organization, 2011. [3] G. Sayer, et al., "Ventricular assist device therapy," Cardiovascular therapeutics, vol. 27, pp. 140-150, 2009. [4] E. A. Rose, et al., "Long-term use of a left ventricular assist device for end-stage heart failure," New England Journal of Medicine, vol. 345, pp. 1435-1443, 2001. [5] M. S. Slaughter, et al., "Advanced heart failure treated with continuous-flow left ventricular assist device," New England Journal of Medicine, vol. 361, pp. 2241-2251, 2009. [6] F. Arabia, et al., "Success rates of long-term circulatory assist devices used currently for bridge to heart transplantation," ASAIO journal (American Society for Artificial Internal Organs: 1992), vol. 42, p. M542, 1996. [7] A. G. Rose, et al., "Partial aortic valve fusion induced by left ventricular assist device," The Annals of thoracic surgery, vol. 70, pp. 1270-1274, 2000. [8] J. H. Connelly, et al., "Acquired commissural fusion of aortic valves in patients with left ventricular assist devices," The Journal of heart and lung transplantation, vol. 22, pp. 1291- 1295, 2003. [9] D. M. Karantonis, et al., "Noninvasive Activity‐based Control of an Implantable Rotary Blood Pump: Comparative Software Simulation Study," Artificial organs, vol. 34, pp. E34- E45, 2010. [10] Intro to the heart. http://antranik.org/intro-to-the-heart/, Accessed: December 2013. [11] E. John, "Guyton and Hall textbook of medical physiology," ed: Philadelphia: Elsevier Saunders, 2011. [12] A. Guyton and J. Hall, "Textbook of medicine physiology," Textbook of medicine physiology, 1996. [13] D. Mohrman and L. Heller, Cardiovascular Physiology. Lange medical book,McGraw-Hill, 7th edition: McGraw Hill Professional, 2006. [14] E. L. Yellin, " Mitral valve motion based on intracardiac dynamics and flow pattern modell " cardiovascular engineering 1981. [15] A. Fauci, et al., "Harrison's principles of internal medicine 14th ed. 1998, 1476 McGraw Hill," San Francisco, CA. [16] D. E. Mohrman and L. J. Heller, "Cardiovascular physiology," 2002. [17] A. M. Katz, Physiology of the Heart: Lippincott Williams & Wilkins, 2010. [18] J. De Hart, et al., "A three-dimensional computational analysis of fluid–structure interaction in the aortic valve," Journal of biomechanics, vol. 36, pp. 103-112, 2003. [19] B. J. Bellhouse and L. Talbot, "The fluid mechanics of the aortic valve," Journal of fluid mechanics, vol. 35, pp. 721-735, 1969. [20] B. Bellhouse and F. Bellhouse, "Fluid mechanics of model normal and stenosed aortic valves," Circulation research, vol. 25, p. 693, 1969. [21] A. P. Yoganathan, et al., "Fluid mechanics of heart valves," Annu. Rev. Biomed. Eng., vol. 6, pp. 331-362, 2004. [22] A. Yoganathan, et al., "Hard Tissue Replacement," The Biomedical Engineering Handbook: Second Edition. Ed. Joseph D. Bronzino, Boca Raton: CRC Press LLC, 2000. [23] C. M. Otto, The Practice of Clinical Echocardiography, third ed. Philadelphia, Pennsylvania W.B. Saunders Company, 1997. 212 [24] C. M. Otto, "Evaluation and management of chronic mitral regurgitation," New England Journal of Medicine, vol. 345, pp. 740-746, 2001. [25] M. S. Sacks, et al., "On the biomechanics of heart valve function," Journal of biomechanics, vol. 42, pp. 1804-1824, 2009. [26] B. Bellhouse, "Velocity and pressure distributions in the aortic valve," Journal of Fluid Mechanics, vol. 37, pp. 587-600, 1969. [27] H. Reul and N. Talukder, "Heart valve mechanics," Quantitative Cardiovascular Studies Clinical and Research Applications of Engineering Principles, pp. 527-564, 1979. [28] WHO. ( Feb, 2013). Cardiovascular diseases (CVDs),World Health Organization Fact sheet No.317. Available: http://www.who.int/mediacentre/factsheets/fs317/en/index.html [29] V. Fuster, Hurst's the heart. New York: New York : McGraw-Hill, Health Professions Div., 2001. [30] B. A. Carabello, "Evolution of the study of left ventricular function: everything old is new again," Circulation, vol. 105, p. 2701, 2002. [31] K. Chatterjee and B. Massie, "Systolic and diastolic heart failure: differences and similarities," Journal of cardiac failure, vol. 13, pp. 569-576, 2007. [32] M. Packer, "Neurohormonal interactions and adaptations in congestive heart failure," Circulation, vol. 77, p. 721, 1988. [33] J. S. Forrester, et al., "Correlative classification of clinical and hemodynamic function after acute myocardial infarction," The American journal of cardiology, vol. 39, pp. 137-145, 1977. [34] E. Braunwald and M. R. Bristow, "Congestive heart failure: fifty years of progress," Circulation, vol. 102, pp. Iv-14-Iv-23, 2000. [35] M. Packer, "Neurohormonal interactions and adaptations in congestive heart failure," Circulation, vol. 77, pp. 721-730, 1988. [36] M. Dolgin, Nomenclature and criteria for diagnosis of diseases of the heart and great vessels: Little, Brown Boston^ eMA MA, 1994. [37] C. H. Selzman, et al., "Surgical therapy for heart failure," Journal of the American College of Surgeons, vol. 203, pp. 226-239, 2006. [38] J. D. Hardy, et al., "Heart transplantation in man: developmental studies and report of a case," Jama, vol. 188, pp. 1132-1140, 1964. [39] R. C. Starling and J. B. Young, "Surgical therapy for dilated cardiomyopathy," Cardiology clinics, vol. 16, pp. 727-737, 1998. [40] D. D. McManus, et al., "Prognostic value of left ventricular end-systolic volume index as a predictor of heart failure hospitalization in stable coronary artery disease: data from the Heart and Soul Study," Journal of the American Society of Echocardiography, vol. 22, pp. 190-197, 2009. [41] S. Vandenberghe, "Modeling the interaction between cardiac assist devices and the left ventricle," 2004. [42] E. Braunwald and M. R. Bristow, "Congestive heart failure: fifty years of progress," Circulation, vol. 102, pp. IV-14, 2000. [43] D. M. Lloyd-Jones, et al., "Lifetime risk for developing congestive heart failure," Circulation, vol. 106, pp. 3068-3072, 2002. [44] M. Á. Fernández and M. Moubachir, "A Newton method using exact Jacobians for solving fluid–structure coupling," Computers & Structures, vol. 83, pp. 127-142, 2005. [45] E. A. Rose, et al., "The REMATCH trial: rationale, design, and end points," The Annals of thoracic surgery, vol. 67, p. 723, 1999. [46] G. V. Letsou, et al., "Continuous axial-flow left ventricular assist device (Jarvik 2000) maintains kidney and liver perfusion for up to 6 months," The Annals of thoracic surgery, vol. 76, p. 1167, 2003. 213 [47] K. L. Baughman and J. A. Jarcho, "Bridge to life—cardiac mechanical support," New England Journal of Medicine, vol. 357, pp. 846-849, 2007. [48] P. A. Watterson, et al., "VentrAssist hydrodynamically suspended, open, centrifugal blood pump," Artificial Organs, vol. 24, pp. 475-477, 2000. [49] G. V. Letsou, et al., "Is native aortic valve commissural fusion in patients with long-term left ventricular assist devices associated with clinically important aortic insufficiency?," The Journal of heart and lung transplantation, vol. 25, pp. 395-399, 2006. [50] J. Cowger, et al., "The Development of Aortic Insufficiency in Left Ventricular Assist Device- Supported PatientsClinical Perspective," Circulation: Heart Failure, vol. 3, pp. 668-674, 2010. [51] L. E. Samuels, et al., "Insufficiency of the native aortic valve and left ventricular assist system inflow valve after support with an implantable left ventricular assist system: signs, symptoms, and concerns," Journal of thoracic and cardiovascular surgery, vol. 122, pp. 380- 381, 2001. [52] A. S. Bryant, et al., "Native aortic valve insufficiency in patients with left ventricular assist devices," The Annals of thoracic surgery, vol. 81, pp. e6-e8, 2006. [53] D. Haghi, et al., "Aortic regurgitation during left ventricular assist device support," The Journal of Heart and Lung Transplantation, vol. 26, pp. 1220-1221, 2007. [54] J. O. Mudd, et al., "Fusion of aortic valve commissures in patients supported by a continuous axial flow left ventricular assist device," Journal of Heart and Lung Transplantation, vol. 27, pp. 1269-1274, 2008. [55] A. Funakubo, et al., "Flow rate and pressure head estimation in a centrifugal blood pump," Artificial Organs, vol. 26, pp. 985-990, 2002. [56] H. Schima, et al., "Noninvasive monitoring of rotary blood pumps: necessity, possibilities, and limitations," Artificial Organs, vol. 16, pp. 195-202, 1992. [57] Y. C. Yu, et al., "Parameter estimation in cardiovascular models," pp. 1433-1434 vol. 2. [58] E. Lim, et al., "Parameter-Optimized Model of Cardiovascular–Rotary Blood Pump Interactions," Biomedical Engineering, IEEE Transactions on, vol. 57, pp. 254-266, 2010. [59] Y. C. Yu, et al., "Mathematical model of pulsatile blood pump for LVAS control," 1997, pp. 3709-3713 vol. 6. [60] T. Kitamura, et al., "Physical Model Based Indirect Measurements of Blood Pressure and Flow Using a Centrifugal Pump," Artificial Organs, vol. 24, pp. 589-593, 2000. [61] Y. Wakisaka, et al., "Establishment of flow estimation for an implantable centrifugal blood pump," ASAIO Journal, vol. 43, p. M663, 1997. [62] Y. Wakisaka, et al., "Noninvasive pump flow estimation of a centrifugal blood pump," Artificial Organs, vol. 21, pp. 651-654, 1997. [63] A. Tanaka, et al., "In vivo evaluation of pressure head and flow rate estimation in a continuous-flow artificial heart," 2001, pp. 3060-3063 vol. 3. [64] M. Yoshizawa, et al., "Sensorless estimation of pressure head and flow of a continuous flow artificial heart based on input power and rotational speed," ASAIO Journal, vol. 48, p. 443, 2002. [65] N. Malagutti, et al., "Noninvasive average flow estimation for an implantable rotary blood pump: a new algorithm incorporating the role of blood viscosity," Artificial Organs, vol. 31, pp. 45-52, 2007. [66] C. Bertram, "Measurement for implantable rotary blood pumps," Physiological measurement, vol. 26, p. R99, 2005. [67] Y. Wu, et al., "Study of pressure estimation for a human circulatory system with a LVAD," p. 6 pp. [68] P. Ayre, et al., "Non-invasive flow estimation in an implantable rotary blood pump: a study considering non-pulsatile and pulsatile flows," Physiological measurement, vol. 24, p. 179, 2003. 214 [69] E. Lim, et al., "Noninvasive average flow and differential pressure estimation for an implantable rotary blood pump using dimensional analysis," Biomedical Engineering, IEEE Transactions on, vol. 55, pp. 2094-2101, 2008. [70] P. Ayre, et al., "Sensorless flow and head estimation in the VentrAssist rotary blood pump," Artificial organs, vol. 24, pp. 585-588, 2001. [71] A. H. H. AlOmari, et al., "A Dynamical Model for Pulsatile Flow Estimation in a Left Ventricular Assist Device," Biosignals 2009: Proceedings of the International Conference on Bio-Inspired Systems and Signal Processing, pp. 402-405, 2009. [72] A. AlOmari, et al., "Non-invasive estimation and control of inlet pressure in an implantable rotary blood pump for heart failure patients," Physiological measurement, vol. 32, p. 1035, 2011. [73] B. McInnis, et al., "Adaptive control of left ventricular bypass assist devices," Automatic Control, IEEE Transactions on, vol. 30, pp. 322-329, 1985. [74] B. McInnis, et al., "Adaptive control of left ventricular bypass assist devices," Automatic Control, IEEE Transactions on, vol. 30, pp. 322-329, 1985. [75] T. Kitamura, "Left atrial pressure controller design for an artificial heart," Biomedical Engineering, IEEE Transactions on, vol. 37, pp. 164-169, 1990. [76] D. M. Karantonis, et al., "Noninvasive Activity based Control of an Implantable Rotary Blood Pump: Comparative Software Simulation Study," Artificial Organs, vol. 34, pp. E34-E45, 2010. [77] M. Nakamura, et al., "Mixed venous oxygen saturation as a promising parameter for physiologic control of total artificial heart," ASAIO journal, vol. 46, p. 761, 2000. [78] G. A. Giridharan and M. Skliar, "Nonlinear controller for ventricular assist devices," Artificial Organs, vol. 26, pp. 980-984, 2002. [79] E. Lim, et al., "Noninvasive deadbeat control of an implantable rotary blood pump: A simulation study," 2009, pp. 2855-2858. [80] A. H. H. AlOmari, et al., "Modeling and control of an implantable rotary blood pump for heart failure patients," 2010, pp. 3574-3579. [81] A. AlOmari, et al., "Non-invasive estimation of pulsatile flow and differential pressure in an implantable rotary blood pump for heart failure patients," Physiological measurement, vol. 30, p. 371, 2009. [82] C. Farhat, et al., "Load and motion transfer algorithms for fluid/structure interaction problems with non-matching discrete interfaces: momentum and energy conservation, optimal discretization and application to aeroelasticity," Computer methods in applied mechanics and engineering, vol. 157, pp. 95-114, 1998. [83] R. Campbell and E. Paterson, "Fluid–structure interaction analysis of flexible turbomachinery," Journal of Fluids and Structures, vol. 27, pp. 1376-1391, 2011. [84] E. J. Hearn, Mechanics of Materials 2: The mechanics of elastic and plastic deformation of solids and structural materials vol. 2: Butterworth-Heinemann, 1997. [85] J. D. Anderson, Computational fluid dynamics vol. 206: Springer, 1995. [86] D. Taylor and J. Wade, "Pattern of blood flow within the heart: a stable system," Cardiovascular research, vol. 7, pp. 14-21, 1973. [87] W. Y. Kim, et al., "Left ventricular blood flow patterns in normal subjects: a quantitative analysis by three-dimensional magnetic resonance velocity mapping," Journal of the American College of Cardiology, vol. 26, pp. 224-238, 1995. [88] J. D. Thomas and A. E. Weyman, "Numerical modeling of ventricular filling," Annals of biomedical engineering, vol. 20, pp. 19-39, 1992. [89] M. Black, et al., "A three-dimensional analysis of a bioprosthetic heart valve," Journal of biomechanics, vol. 24, pp. 793-801, 1991. [90] K. Chandran, et al., "Stress distribution on the cusps of a polyurethane trileaflet heart valve prosthesis in the closed position," Journal of biomechanics, vol. 24, pp. 385-395, 1991. 215 [91] J. D. Hart, et al., "A three-dimensional analysis of a fibre-reinforced aortic valve prosthesis," Journal of biomechanics, vol. 31, pp. 629-638, 1998. [92] M. Vollkron, et al., "Interaction of the Cardiovascular System with an Implanted Rotary Assist Device: Simulation Study with a Refined Computer Model>," Artificial organs, vol. 26, pp. 349-359, 2002. [93] B. Baccani, et al., "Model and influence of mitral valve opening during the left ventricular filling," Journal of biomechanics, vol. 36, pp. 355-361, 2003. [94] Y. Cheng, et al., "Fluid-structure coupled CFD simulation of the left ventricular flow during filling phase," Annals of biomedical engineering, vol. 33, pp. 567-576, 2005. [95] M. P. Nash and P. J. Hunter, "Computational mechanics of the heart," Journal of elasticity and the physical science of solids, vol. 61, pp. 113-141, 2000. [96] Q. Long, et al., "Subject-specific computational simulation of left ventricular flow based on magnetic resonance imaging," Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine, vol. 222, pp. 475-485, 2008. [97] T. J. Hughes, et al., "Lagrangian-Eulerian finite element formulation for incompressible viscous flows," Computer methods in applied mechanics and engineering, vol. 29, pp. 329- 349, 1981. [98] B. Bellhouse, "Fluid mechanics of a model mitral valve and left ventricle," Cardiovascular research, vol. 6, pp. 199-210, 1972. [99] H. Reul and N. Talukder, "Fluid mechanics of the natural mitral valve," Journal of biomechanics, vol. 14, pp. 361-372, 1981. [100] T. W. Taylor, et al., "Three-dimensional analysis of left ventricular ejection using computational fluid dynamics," Journal of biomechanical engineering, vol. 116, pp. 127- 130, 1994. [101] T. Taylor and T. Yamaguchi, "Realistic three-dimensional left ventricular ejection determined from computational fluid dynamics," Medical engineering & physics, vol. 17, pp. 602-608, 1995. [102] T. N. Jones and D. N. Metaxas, "Patient-specific analysis of left ventricular blood flow," in Medical Image Computing and Computer-Assisted Interventation—MICCAI’98, ed: Springer, 1998, pp. 156-166. [103] J. Vierendeels, et al., "Computer simulation of left ventricular filling flow: impact study on echocardiograms," in Computers in Cardiology, 1999, 1999, pp. 177-180. [104] C. S. Peskin, "Flow patterns around heart valves: a numerical method," Journal of computational physics, vol. 10, pp. 252-271, 1972. [105] C. S. Peskin and D. M. McQueen, "A three-dimensional computational method for blood flow in the heart I. Immersed elastic fibers in a viscous incompressible fluid," Journal of Computational Physics, vol. 81, pp. 372-405, 1989. [106] D. M. McQueen and C. S. Peskin, "A three-dimensional computational method for blood flow in the heart. II. Contractile fibers," Journal of Computational Physics, vol. 82, pp. 289- 297, 1989. [107] D. M. McQueen and C. S. Peskin, "A three-dimensional computer model of the human heart for studying cardiac fluid dynamics," ACM SIGGRAPH Computer Graphics, vol. 34, pp. 56-60, 2000. [108] E. J. Vigmond, et al., "Effect of bundle branch block on cardiac output: a whole heart simulation study," Progress in biophysics and molecular biology, vol. 97, pp. 520-542, 2008. [109] D. M. McQUEEN, et al., "Fluid dynamics of the mitral valve: physiological aspects of a mathematical model," American Journal of Physiology-Heart and Circulatory Physiology, vol. 242, pp. H1095-H1110, 1982. [110] C. S. Peskin, "The fluid dynamics of heart valves: experimental, theoretical, and computational methods," Annual review of fluid mechanics, vol. 14, pp. 235-259, 1982. 216 [111] C. C. Vesier, et al., "A three-dimensional computational model of a thin-walled left ventricle," in Proceedings of the 1992 ACM/IEEE conference on Supercomputing, 1992, pp. 73-82. [112] J. D. Lemmon and A. P. Yoganathan, "Computational modeling of left heart diastolic function: examination of ventricular dysfunction," Journal of biomechanical engineering, vol. 122, pp. 297-303, 2000. [113] P. J. Kilner, et al., "Asymmetric redirection of flow through the heart," Nature, vol. 404, pp. 759-761, 2000. [114] N. R. Saber, et al., "Computational flow modeling of the left ventricle based on in vivo MRI data: initial experience," Annals of biomedical engineering, vol. 29, pp. 275-283, 2001. [115] T. Ebbers, et al., "Noninvasive measurement of time-varying three-dimensional relative pressure fields within the human heart," Journal of biomechanical engineering, vol. 124, pp. 288-293, 2002. [116] T. Akimoto, et al., "Rotary blood pump flow spontaneously increases during exercise under constant pump speed: results of a chronic study," Artificial organs, vol. 23, pp. 797-801, 1999. [117] M. Nakamura, et al., "Computational analysis of blood flow in an integrated model of the left ventricle and the aorta," Journal of biomechanical engineering, vol. 128, pp. 837-843, 2006. [118] Q. Long, et al., "The influence of inflow boundary conditions on intra left ventricle flow predictions," Journal of biomechanical engineering, vol. 125, pp. 922-927, 2003. [119] N. Forsythe and J.-D. Mueller, "Validation of a fluid–structure interaction model for a bileaflet mechanical heart valve," International Journal of Computational Fluid Dynamics, vol. 22, pp. 541-553, 2008. [120] S. Krittian, et al., "Partitioned fluid–solid coupling for cardiovascular blood flow," Annals of biomedical engineering, vol. 38, pp. 1426-1441, 2010. [121] I. Borazjani, et al., "Curvilinear immersed boundary method for simulating fluid structure interaction with complex 3D rigid bodies," Journal of Computational physics, vol. 227, pp. 7587-7620, 2008. [122] B. Chahboune and J. Crolet, "Numerical simulation of the blood-wall interaction in the human left ventricle," The European Physical Journal Applied Physics, vol. 2, pp. 291-297, 1998. [123] T. Dunne, et al., "Numerical simulation of fluid-structure interaction based on monolithic variational formulations," Fundamental trends in fluid–structure interaction, Contemp. Chall. Math. Fluid Dyn. Appl, vol. 1, pp. 1-75, 2010. [124] J. Vierendeels, et al., "Computer simulation of intraventricular flow and pressure gradients during diastole," Journal of biomechanical engineering, vol. 122, pp. 667-674, 2000. [125] K. Riemselagh, et al., "Two-Dimensional Incompressible Navier-Stokes Calculations in Complex-Shaped Moving Domains," Journal of Engineering Mathematics, vol. 34, pp. 57-73, 1998/07/01 1998. [126] J. De Hart, et al., "A two-dimensional fluid–structure interaction model of the aortic value," Journal of biomechanics, vol. 33, pp. 1079-1088, 2000. [127] J. D. Hart, "Fluid-structure interaction in the aortic heart valve: a three dimensional computational analysis," Thesis, Technische Universiteit Eindhoven, 2002. [128] J. De Hart, et al., "A computational fluid-structure interaction analysis of a fiber-reinforced stentless aortic valve," Journal of biomechanics, vol. 36, pp. 699-712, 2003. [129] P. J. Hunter, et al., "Modeling total heart function," Annual review of biomedical engineering, vol. 5, pp. 147-177, 2003. [130] R. Van Loon, et al., "A combined fictitious domain/adaptive meshing method for fluid– structure interaction in heart valves," International journal for numerical methods in fluids, vol. 46, pp. 533-544, 2004. 217 [131] J. Stijnen, et al., "Evaluation of a fictitious domain method for predicting dynamic response of mechanical heart valves," Journal of Fluids and Structures, vol. 19, pp. 835-850, 2004. [132] H. Watanabe, et al., "Computer simulation of blood flow, left ventricular wall motion and their interrelationship by fluid-structure interaction finite element method," JSME international journal. Series C, Mechanical systems, machine elements and manufacturing, vol. 45, pp. 1003-1012, 2002. [133] H. Watanabe, et al., "Multiphysics simulation of left ventricular filling dynamics using fluid- structure interaction finite element method," Biophysical journal, vol. 87, pp. 2074-2085, 2004. [134] D. L. Timms, et al., "A compact mock circulation loop for the in vitro testing of cardiovascular devices," Artificial organs, vol. 35, pp. 384-391, 2011. [135] Y. Shi, et al., "Numerical simulation of cardiovascular dynamics with different types of VAD assistance," Journal of biomechanics, vol. 40, pp. 2919-2933, 2007. [136] W. A. Smith, et al., "System analysis of the flow/pressure response of rotodynamic blood pumps," Artificial organs, vol. 23, pp. 947-955, 1999. [137] J. B. A. M. Horsten, "On the analysis of moving heart valves: a numerical fluid-structure interaction model," 1990. [138] D. Nordsletten, et al., "Fluid–solid coupling for the investigation of diastolic and systolic human left ventricular function," International Journal for Numerical Methods in Biomedical Engineering, vol. 27, pp. 1017-1039, 2010. [139] K. Dumont, et al., "Comparison of the hemodynamic and thrombogenic performance of two bileaflet mechanical heart valves using a CFD/FSI model," Journal of biomechanical engineering, vol. 129, pp. 558-565, 2007. [140] R. van Loon, et al., "A three-dimensional fluid–structure interaction method for heart valve modelling," Comptes Rendus Mecanique, vol. 333, pp. 856-866, 2005. [141] J. Stijnen, et al., "Computational analysis of ventricular valve–valve interaction: Influence of flow conditions," International Journal of Computational Fluid Dynamics, vol. 23, pp. 609- 622, 2009. [142] C. S. Peskin, "Numerical analysis of blood flow in the heart," Journal of computational physics, vol. 25, pp. 220-252, 1977. [143] T. Chung, Computational fluid dynamics: Cambridge university press, 2010. [144] R. Cheng, et al., "Three-dimensional fluid-structure interaction simulation of bileaflet mechanical heart valve flow dynamics," Annals of biomedical engineering, vol. 32, pp. 1471-1483, 2004. [145] D. M. Espino, et al., "Simulation of Blood Flow through the Mitral Valve of the Heart: A Fluid Structure Interaction Model," in Proc. COMSOL Users Conference, 2006. [146] J. Donea, et al., "An arbitrary Lagrangian-Eulerian finite element method for transient dynamic fluid-structure interactions," Computer Methods in Applied Mechanics and Engineering, vol. 33, pp. 689-723, 1982. [147] N. Westerhof, et al., "An artificial arterial system for pumping hearts," Journal of Applied Physiology, vol. 31, pp. 776-781, 1971. [148] M. Danielsen, "Modeling of feedback mechanisms which control the heart function in a view to an implementation in cardiovascular models," 1998. [149] E. Lim, et al., "A dynamic lumped parameter model of the left ventricular assisted circulation," in Engineering in Medicine and Biology Society, 2007. EMBS 2007. 29th Annual International Conference of the IEEE, 2007, pp. 3990-3993. [150] J. Bogaert and D. S, Cardiac Anatomy. In:clinical MRI: Springer-Verlag, 2005. [151] C. Hirt, et al., "An arbitrary Lagrangian-Eulerian computing method for all flow speeds," Journal of Computational Physics, vol. 14, pp. 227-253, 1974. 218 [152] O. Schenk and K. Gärtner, "Solving unsymmetric sparse systems of linear equations with PARDISO," Future Generation Computer Systems, vol. 20, pp. 475-487, 2004. [153] F. Yilmaz and M. Y. Gundogdu, "A critical review on blood flow in large arteries; relevance to blood rheology, viscosity models, and physiologic conditions," Korea-Australia Rheology Journal, vol. 20, pp. 197-211, 2008. [154] V. Rao, et al., "Surgical management of valvular disease in patients requiring left ventricular assist device support," The Annals of thoracic surgery, vol. 71, pp. 1448-1453, 2001. [155] E. Gologorsky, et al., "Monitoring of aortic valve opening and systolic aortic insufficiency in optimization of continuous-flow left ventricular assist device settings," J Cardiothorac Vasc Anesth, vol. 26, pp. 1063-6, Dec 2012. [156] T. J. Myers, et al., "Assessment of arterial blood pressure during support with an axial flow left ventricular assist device," The Journal of Heart and Lung Transplantation, vol. 28, pp. 423-427, 2009. [157] M. Granegger, et al., "Assessment of Aortic Valve Opening During Rotary Blood Pump Support Using Pump Signals," Artificial organs, 2013. [158] K. A. Alonazi, et al., "Modeling aortic valve closure under the action of a ventricular assist device," in Engineering in Medicine and Biology Society (EMBC), 2013 35th Annual International Conference of the IEEE, 2013, pp. 679-682. [159] J. T. Butcher, et al., "Review—Mechanobiology of the aortic heart valve," J Heart Valve Dis, vol. 17, 2008. [160] Y. Naka, et al., "Novel technique to repair type A acute aortic dissection in patients with a left ventricular assist device," The Annals of thoracic surgery, vol. 72, pp. 1403-1404, 2001. [161] C. M. Feldman, et al., "Management of aortic insufficiency with continuous flow left ventricular assist devices: bioprosthetic valve replacement," The Journal of heart and lung transplantation: the official publication of the International Society for Heart Transplantation, vol. 25, p. 1410, 2006. [162] D. Eng and A. Dubovoy, High Left Ventricular Vent Return After Left and Right Ventricular Assist Device Placement in a Patient With a Mechanical Aortic Valve, 2012. [163] A. G. Rose, et al., "Total left ventricular outflow tract obstruction due to left ventricular assist device-induced sub-aortic thrombosis in 2 patients with aortic valve bioprosthesis," The Journal of heart and lung transplantation: the official publication of the International Society for Heart Transplantation, vol. 22, p. 594, 2003. [164] K. D. Aaronson, et al., "Patient selection for left ventricular assist device therapy," The Annals of thoracic surgery, vol. 75, pp. S29-S35, 2003. [165] E. Gologorsky, et al., "Monitoring of Aortic Valve Opening and Systolic Aortic Insufficiency in Optimization of Continuous-Flow Left Ventricular Assist Device Settings," Journal of cardiothoracic and vascular anesthesia, 2012. [166] H. Watanabe, et al., "Computer simulation of blood flow, left ventricular wall motion and their interrelationship by fluid-structure interaction finite element method," JSME International Journal Series C, vol. 45, pp. 1003-1012, 2002. [167] M. McCormick, et al., "Simulating Left Ventricular Fluid-Solid Mechanics through the Cardiac Cycle under LVAD Support," Journal of Computational Physics, 2012. [168] M. McCormick, et al., "Modelling left ventricular function under assist device support," International Journal for Numerical Methods in Biomedical Engineering, vol. 27, pp. 1073- 1095, 2011. [169] A. G. Rose, et al., "Partial aortic valve fusion induced by left ventricular assist device," The Annals of thoracic surgery, vol. 70, p. 1270, 2000. [170] K. A. Alonazi, et al., "Simulation of motor current waveform as an index for aortic valve open-close condition during ventricular support," in Engineering in Medicine and Biology Society (EMBC), 2014 36th Annual International Conference of the IEEE, 2014, pp. 3013- 3016. 219 [171] R. Rasalingam, et al., "Transthoracic echocardiographic assessment of continuous-flow left ventricular assist devices," Journal of the American Society of Echocardiography, vol. 24, pp. 135-148, 2011. [172] S. J. Park, et al., "Left ventricular assist devices as destination therapy: a new look at survival," The Journal of thoracic and cardiovascular surgery, vol. 129, pp. 9-17, 2005. [173] T. Korakianitis and Y. Shi, "Numerical simulation of cardiovascular dynamics with healthy and diseased heart valves," Journal of biomechanics, vol. 39, pp. 1964-1982, 2006. [174] S. Vandenberghe, et al., "Effect of rotary blood pump failure on left ventricular energetics assessed by mathematical modeling," Artificial organs, vol. 26, pp. 1032-1039, 2002. [175] Y. Wu, et al., "Modeling, estimation, and control of human circulatory system with a left ventricular assist device," Control Systems Technology, IEEE Transactions on, vol. 15, pp. 754-767, 2007. [176] Y. Liu, et al., "Design and initial testing of a mock human circulatory loop for left ventricular assist device performance testing," Artificial organs, vol. 29, pp. 341-345, 2005. [177] G. M. Pantalos, et al., "Characterization of an adult mock circulation for testing cardiac support devices," ASAIO journal, vol. 50, pp. 37-46, 2004. [178] D. Timms, et al., "A complete mock circulation loop for the evaluation of left, right, and biventricular assist devices," Artificial organs, vol. 29, pp. 564-572, 2005. [179] A. H. Goldstein, et al., "Partial support with a centrifugal left ventricular assist device reduces myocardial oxygen consumption in chronic, ischemic heart failure," Journal of cardiac failure, vol. 11, pp. 142-151, 2005. [180] M. C. Deng, et al., "Effects of exercise during long-term support with a left ventricular assist device," Circulation, vol. 97, pp. 1212-1213, 1998. [181] H. Schima, et al., "First clinical experience with an automatic control system for rotary blood pumps during ergometry and right-heart catheterization," The Journal of heart and lung transplantation, vol. 25, pp. 167-173, 2006. [182] P. Ayre, et al., "Identifying physiologically significant pumping state transitions in implantable rotary blood pumps used as left ventricular assist devices: an in-vivo study," in Engineering in Medicine and Biology Society, 2001. Proceedings of the 23rd Annual International Conference of the IEEE, 2001, pp. 445-448. [183] D. M. Karantonis, et al., "Identification and classification of physiologically significant pumping states in an implantable rotary blood pump," Artificial Organs, vol. 30, pp. 671- 679, 2006. [184] S. C. Ng, et al., "Evaluation of suction detection during different pumping states in an implantable rotary blood pump," Artificial organs, vol. 37, pp. E145-E154, 2013. [185] K. W. Gwak, et al., "Safety‐Enhanced Optimal Control of Turbodynamic Blood Pumps," Artificial organs, vol. 35, pp. 725-732, 2011. [186] R. M. Lazar, et al., "Neurological events during long-term mechanical circulatory support for heart failure the randomized evaluation of mechanical assistance for the treatment of congestive heart failure (REMATCH) Experience," Circulation, vol. 109, pp. 2423-2427, 2004. [187] A. G. Rose and S. J. Park, "Pathology in patients with ventricular assist devices: a study of 21 autopsies, 24 ventricular apical core biopsies and 24 explanted hearts," Cardiovascular Pathology, vol. 14, pp. 19-23, 2005. [188] D. M. Karantonis, et al., "Noninvasive detection of suction in an implantable rotary blood pump using neural networks," International Journal of Computational Intelligence and Applications, vol. 7, pp. 237-247, 2008. [189] D. M. Karantonis, et al., "Classification of physiologically significant pumping states in an implantable rotary blood pump: patient trial results," ASAIO Journal, vol. 53, pp. 617-622, 2007. 220 [190] C. J. Bishop, et al., "A novel non-invasive method to assess aortic valve opening in HeartMate II left ventricular assist device patients using a modified Karhunen-Loève transformation," The Journal of Heart and Lung Transplantation, vol. 29, pp. 27-31, 2010. [191] M. Granegger, et al., "Assessment of the aortic valve opening during rotary blood pump support," ASAIO J, vol. 57, p. 75, 2011. [192] G. Endo, et al., "The meaning of the turning point of the index of motor current amplitude curve in controlling a continuous flow pump or evaluation of left ventricular function," Artificial organs, vol. 27, pp. 272-276, 2003. [193] A. Yuhki, et al., "Detection of suction and regurgitation of the implantable centrifugal pump based on the motor current waveform analysis and its application to optimization of pump flow," ARTIFICIAL ORGANS-OHIO-, vol. 23, pp. 532-537, 1999. [194] M. Oshikawa, et al., "Sensorless controlling method for a continuous flow left ventricular assist device," Artificial organs, vol. 24, pp. 600-605, 2000. [195] C. S. Peskin and D. M. McQUEEN, "Mechanical equilibrium determines the fractal fiber architecture of aortic heart valve leaflets," American journal of physiology, vol. 266, pp. H319-H319, 1994. [196] R. De Paulis, et al., "Opening and closing characteristics of the aortic valve after valve- sparing procedures using a new aortic root conduit," The Annals of thoracic surgery, vol. 72, pp. 487-494, 2001. [197] R. John, et al., "Aortic valve pathophysiology during left ventricular assist device support," The Journal of Heart and Lung Transplantation, vol. 29, pp. 1321-1329, 2010. [198] D. L. Eckberg and P. Sleight, Human baroreflexes in health and disease: Clarendon Press Oxford, 1992. [199] A. W. Hall, et al., "Physiologic control of cardiac assist devices," Artificial organs, vol. 20, pp. 271-275, 1996. [200] M. Granegger, et al., "Assessment of Aortic Valve Opening During Rotary Blood Pump Support Using Pump Signals," Artificial organs, vol. 38, pp. 290-297, 2014. [201] H. L. Ooi, et al., "Robust Aortic Valve Non‐Opening Detection for Different Cardiac Conditions," Artificial organs, vol. 38, pp. E57-E67, 2014. [202] D. M. Karantonis, et al., "Automated non-invasive detection of pumping states in an implantable rotary blood pump," in Engineering in Medicine and Biology Society, 2006. EMBS'06. 28th Annual International Conference of the IEEE, 2006, pp. 5386-5389. [203] M. Moayeri and G. Zendehbudi, "Effects of elastic property of the wall on flow characteristics through arterial stenoses," Journal of biomechanics, vol. 36, pp. 525-535, 2003. [204] M. Hatano, et al., "Less frequent opening of the aortic valve and a continuous flow pump are risk factors for postoperative onset of aortic insufficiency in patients with a left ventricular assist device," Circulation journal: official journal of the Japanese Circulation Society, vol. 75, p. 1147, 2011. [205] D. Bennett, et al., "Echocardiographic left ventricular dimensions in pressure and volume overload. Their use in assessing aortic stenosis," British heart journal, vol. 37, pp. 971-977, 1975. [206] S. Choi, et al., "Hemodynamic controller for left ventricular assist device based on pulsatility ratio," Artificial organs, vol. 31, pp. 114-125, 2007. [207] S.-H. Jansen-Park, et al., "A monitoring and physiological control system for determining aortic valve closing with a ventricular assist device," European Journal of Cardio-Thoracic Surgery, p. ezu006, 2014. [208] D. Kikugawa, "Evaluation of cardiac function during left ventricular assist by a centrifugal blood pump," Artificial organs, vol. 24, pp. 632-635, 2000. [209] B. Meyns, et al., "Miniaturized implantable rotary blood pump in atrial-aortic position supports and unloads the failing heart," Vascular, vol. 6, pp. 288-295, 1998. 221 222