<<

Berger et al., J Neurobiol Physiol 2021; Journal of Neurobiology and Physiology 3(1):10-15.

Short Communication

The potential of computational fluid dynamics simulations of airflow in the

Berger M1,2, Pillei M1,3, Freysinger W2*

1Department of Environmental, Process Abstract & Energy Engineering, MCI – The Entrepreneurial School, Austria Computational Fluid Dynamics (CFD) is a well-established and accepted tool for simulation and prediction of complex physical phenomena e.g., in combustion, aerodynamics or blood circulation. 2University Hospital of Recently CFD has entered the medical field due to the readily available high computational power of Otorhinolaryngology, Medical current graphics processing units, GPUs. Efficient numerical codes, commercial or open source, are University Innsbruck, Austria available now. Thus, a wide range of medical themes is available for CFD almost in real-time in the 3Department of Fluid Mechanics, medical environment now. Friedrich-Alexander University The available methods are on the point of reaching a usability status ready for everyday clinical use Erlangen-Nuremberg, Germany as a potential medical decision support system, provided adherence to the appropriate patient data *Author for correspondence: protection rules and proper certification as a medical device. Email: wolfgang.freysinger@i-med. ac.at This contribution outlines the current range of activities in our clinic in the field of Lattice-Boltzmann CFD based on CT and / or MR imagery and flashlights the following three areas: simulating the effect Received date: November 05, 2020 of nasal stents on , predicting clinical Rhinomanometry and Rhinometry, and the numerical Accepted date: February 23, 2021 estimation of resection volumes for surgery to improve nasal breathing. The set of required methods is outlined, the three areas are discussed and summarized. The results presented are very preliminary insights into current research and need to be subject to further detailed evaluation, of course. Copyright: © 2021 Berger M, et al. This The results presented are very promising first steps in the direction of both digital patient diagnosis and is an open-access article distributed therapy prediction. However, thorough and in-depth evaluation of the methods themselves and proper under the terms of the Creative clinical studies to assess the merits of the technology are still required. Commons Attribution License, which permits unrestricted use, distribution, Keywords: Nasal airflow, Laser Doppler anemometry, Computational fluid dynamics, Breathing and reproduction in any medium, simulation, Nasal airflow optimization, Rhinomanometry, Surgery, Simulation, Diagnosis prediction provided the original author and source are credited. Abbreviations 3D: Three-dimensional; CFD: Computational Fluid Dynamics; CT: Computer Tomography; HPDR: High Pressure Drop Region; MRI: Magnetic Resonance Imaging; CT: Computed Tomography; LB: Lattice Boltzmann; LDA: Laser Doppler Anemometry

Introduction Numerical prediction of complex physical phenomena in the medical domain almost always implies complex geometries, boundary conditions and challenging physical questions. Paradigmatically, nasal airflow is of interest as this might be a tool to improve the current success rate of and sinus surgery [1,2] as patient satisfaction studies suggest [3]. According to current clinical guidelines preoperative planning of surgery of the nasal cavity is, in addition to 2D/3D radiologic imaging and nasal endoscopy more and more including additional Rhinomanometry and data [4]. The success of surgical intervention in general and specifically in optimizing nasal breathing (i.e., airflow), strongly depends on the surgeon’s experience. It might thus be worthy to investigate whether digital aids could help doctors to improve this situation. Citation: Berger M, Pillei M, Freysinger Moreover, consequent use of available digital patient data eventually could help advancing the W. The potential of computational fluid current state of medical practice. dynamics simulations of airflow in the Lattice Boltzmann (LB) fluid flow simulations are simple and fast [5] to simulate nasal airflow. nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15. In the microscopic description state of a gas/fluid by a Newtonian fluid model every particle is assigned position and velocity. As this is impossible to realize in numerical simulations the

This article is originally published by ProBiologist LLC., and is freely available at probiologists.com

J Neurobiol Physiol 2021; 3(1):10-15. 10 Citation: Berger M, Pillei M, Freysinger W. The potential of computational fluid dynamics simulations of airflow in the nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15.

computational model builds on statistical considerations on base of spatial resolution 0.3 ×0.3 ×0.3 mm³, 536 × 536 × 440) velocity distribution functions. data As with any simulation tool, LB simulations need to be validated The study was conducted in accordance with local ethical against ground truth. This was done in a thorough investigation where simulations and Laser Doppler Anemometry, LDA, of a guidelines as stipulated by the 7th revision of the Declaration physical (3D printed) and a 3D reconstruction of a 3D-CT image of Helsinki. No ethics committee approval for this anonymized stack was performed [6]. A simplified phantom and an anonymous retrospective study was needed. All anonymous data used in the patient CT data set were segmented to show only tissue and air [7] computations originate from patients who voluntarily agreed to of which the tissue part was 3D printed [8]. This allowed a realistic anonymized use of their data for further scientific purposes; some reproduction of the paranasal sinuses, especially the main nasal data were acquired at different times and therefore not all physiologic cavity that is providing the predominant nasal airflow for breathing states of relevance (e.g., status of the nasal cycle) are known. For one at the zones of the inferior/middle turbinates. LDA measurements patient CT and MR data, Rhinomanometry and Rhinometry were and CFD simulations were found to be in good agreement. It is thus available to allow a comparison of clinical and computed diagnoses. justified to use the LB simulation tool in this context. CT data were acquired with swollen nasal mucosa (i.e., with present infection or acute allergic state); MR data were acquired in a normal A novel medical device for treating snoring conditions is finding condition of the nasal mucosa. The actual state of the nasal cycle was more and more patient acceptance [9]; this nasal stent made from not taken into consideration. Nitinol widens the nasal air passage between lower and middle turbinate during overnight use. Currently daytime use for sports The computational time of one simulation was about 4 minutes activities of this device seems to be a new field of use. Available on a CPU XEON (E5 1650 V3) with twelve threads and 32 GB anonymous MR imagery were used in CFD simulations to assess RAM (4 × 8 GB DDR4-2133) and a NVIDIA RTX 2080 Ti the physical effects of these stents on nasal airflow with and without graphics card (GPU). The fluid flow simulations were performed on the nasal stent. In addition, the prediction tool for surgery was run GPU. The Rhinomanometry curve was simulated within less than on these data sets. 3 hours. For an exceptional single case CT and MR imagery, and Preparation of radiologic 3D data Rhinomanometry and Rhinometry data, all anonymous, allowed All data sets were thresholded and segmented [7,10] to show an initial test run of the LB simulation framework for predicting air and tissue spaces only. At the in- and outlets (i.e., nostrils and clinical diagnoses. nasopharynx) cuboids (60 × 40 × 20 mm³) were added as proper The basic idea behind all these approaches is to use CFD to boundary volumes for the LB simulations. The resulting mesh form calculate streamlines and pressure drops with given boundary this volume was obtained with [11] and was saved in .stl format for conditions of pressure, flow, and anatomy. Analysis of the effects of the simulations. Nasal airflow cross-sections were obtained from the the Alaxo Stents and the prediction of surgical resection volumes segmented data in coronal views in steps of 5 mm from the nostrils relies on the determination of certain high-pressure-drop regions in posterior direction with 3D Slicer [10] and were compared to the along streamlines (see Materials and Methods section), whereas the data of acoustic Rhinometry. simulations of Rhinomanometry and Rhinometry basically employ evaluating the volume flux and the cross-sections of the paranasal LB simulation and optimization sinuses in coronal slice orientations, respectively. Inhalation was simulated with a constant air flow rate of 0.6 The investigations were predominately based on anonymous l/s for reasons of computational stability with Sailfish CFD [5]. CT datasets, being the clinical standard imagery; CT and MR Inlet and outlet were set to ambient pressure, respectively, and at imaging were used for predicting diagnostic Rhinomanometry and the outlet a flow boundary condition was set. Simulations started Rhinometry data; MR data were used for simulating the Alaxo stents. with velocity increasing from zero until a quasi-stationary solution The nature of the 3D imaging modality used to generate the three- was achieved. The physical situation gives a Reynolds number (Re) dimensional volumes to run the CFD simulations is defined by the indicating turbulences that were handled with the LB large eddy clinical availability or the diagnostic requirements; other than that, simulation (LES) option of Sailfish. CFD simulations served to simulation does not put constraints on the medical imagery used. find high-pressure drop regions in the three-dimensional flow field, The three-dimensional patient imagery has to allow the generation of HPDRs, to indicate zones of surgical removal of tissue to optimize meshes, three-dimensional surfaces, of the boundary between air and nasal breathing; air-flow cross-section was increased until less than tissue only. Some minor adjustments in the specific segmentation 5 HPDRs were found [Berger et al, JCARS under review]. From approaches might become necessary for different medical imaging the initial set of HPDRs users manually selected the ones most modalities. appropriate to clinical requirements. Cases of septum deviations served as examples in this investigation. The data with nasal splints Material and Methods in place were simulated accordingly. Stent MR data (Siemens, T1 mpr_ns_sag_fast, Echo Train In order to simulate clinical Rhinomanometry data for inspiration Length 1, Flip angle 15, Inversion time 760 ms, 0.488 × 0.488 × 1 and expiration of both nostrils air flow was increased from 0 to 600 mm³, 512 × 512 × 103). MR data comparison to Rhinomanometry ml/s in increments of 50 ml/s. In the segmented model one nostril (t1_mpr_ns_tra, 0.9375 × 0.9375 × 1.2 mm³, 208 × 256 × 128). was blocked to resemble clinical Rhinomanometry. For comparing CT data comparison to Rhinomanometry (Siemens with simulations clinical Rhinomanometry data were approximated SOMATOM, beam current 88 mA, convolution kernel H30 s, by the Bernoulli equation [12].

J Neurobiol Physiol 2021; 3(1):10-15. 11 Citation: Berger M, Pillei M, Freysinger W. The potential of computational fluid dynamics simulations of airflow in the nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15.

Results Figure 1 shows that in the data sets without stent HPDRs are identified at locations where the nasal stent is actually positioned, Table 1 shows that nasal stents reduce the pressure drops red points. For patient 2 simulations with and without stent gave the between inlet and outlet cuboids significantly. The optimizations same results; however, a 148 Pa pressure drop is most likely due to clearly show that applying a nasal stent leads to results that compare the fact that the air-flow cross-section at the nasopharynx was much well to simulated surgery results with very small pressure drops. smaller without stent indicating a different physiologic / anatomic Details in Berger et al, JCARS under review. Table 2 shows that the status. geometry of the nasal cavity with applied stent would require much less surgical optimization than the one indicated by the amount of In Figure 2 the corresponding estimated resection volume tissue to be resected in the simulations. was determined. Where the nasal stent was in place, no further

Pressure drops [Pa] without stent, CFD with stent, with stent, CFD optimized without stent, from CFD optimized geometry from CFD geometry Patient 1 371 11 200 16 Patient 2 148 12 148 10 Patient 3 241 13 69 16

Table 1: Simulated pressure drops between inlet/outlet of the nasal cavity with and without nasal stent and with simulated surgery.

Resection Volume [cm³] without stent with stent Patient 1 14.6 10.2 Patient 2 6.89 6.52 Patient 3 12.52 5.14

Table 2: Estimated resection volume for the three patients indicates that for an applied stent the geometry hardly needs surgical “optimization”.

Figure 1: Segmented nasal cavity with found high pressure drop regions (HPDRs) – red points. With applied stents the number of HPDRs is much smaller.

J Neurobiol Physiol 2021; 3(1):10-15. 12 Citation: Berger M, Pillei M, Freysinger W. The potential of computational fluid dynamics simulations of airflow in the nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15.

optimization was proposed, indicating that a nasal stent “optimized” approximated Rhinomanometry data and simulation was 60.25 Pa nasal breathing. Table 2 shows the estimated corresponding and 148.43 Pa, respectively. Detumesced right and left sides differed calculated resection volumes. by 8.40 Pa and 58.23 Pa, respectively [13]. The right panel of Figure Figure 3 shows the comparison simulated and measured acoustic 3 right shows the segmentation results and acoustic Rhinometry Rhinometry and Rhinomanometry data [13]. The left panel shows data. The airflow cross-section RMSE of the left cavity, not swollen the LB simulation results (points) superimposed to Rhinometry data was 0.81 cm², swollen 2.32 cm²; right: not swollen 0.63 cm², swollen (lines). The root mean square error (RMSE) of the pressure drops 1.54 cm². In all investigations, simulation and segmentation based between swollen left and right nasal cavity between the Bernoulli on MRI datasets are in better accordance with measurements than

Figure 2: Segmented nasal cavity and estimated resection volume (red).

Figure 3: Left: Comparison of Rhinomanometry measurement data with LB simulation results based on CT and MRI dataset of the same patient. On the abscissa the pressure drop [Pa] between nostril and throat is shown, on the ordinate the flow rate ( ) ml/s is depicted. Yellow lines are Bernoulli approximations [13]. Right: Comparison of acoustic Rhinometry measurement data with segmentation results of the same patient. The distance in cm is the acoustic Rhinometry distance from the measurement device, at the simulations those are in coronal direction from nostril.

J Neurobiol Physiol 2021; 3(1):10-15. 13 Citation: Berger M, Pillei M, Freysinger W. The potential of computational fluid dynamics simulations of airflow in the nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15.

Figure 4: Optimization results of the CT data of the patient with Rhinomanometry and Rhinometry data. Coronal airflow cross-sections. Background: CT image, Magenta: suggested resection volume. investigations based on CT datasets. In a recent study it was found Conclusion that LB simulations provided excellent correlation (r=0.97, p<0.001) of measured and simulated Rhinomanometry and Rhinometry data LB fluid flow simulations have some potential in the preoperative [Berger et al JCARS, accepted]. planning process to improve the surgical outcome of surgeries to improve nasal breathing. Compared to CT, MRI is considered to be Figure 4 shows optimization results with the LB method based on harmless for patients and similarly useful for LB CFD simulations. the CT dataset of the patient for whom CT, MR, Rhinomanometry Segmentation of MR data, however, is somewhat more difficult than and acoustic Rhinometry data were available. Coronal slices close to segmentation of CT data. the nasal valve are depicted. Magenta shows the calculated resection volume. Conflicts of Interest Discussion None of the authors declare any conflict of interest. This short communication shows extensions to the work [6], References and to our current work (partly still under review). All previous 1. Gillman GS, Egloff AM, Rivera‐Serrano CM. Revision septoplasty: investigations were performed with CT datasets that allowed easy A prospective disease‐specific outcome study. The Laryngoscope. automated segmentation of air and tissue spaces. MRI imagery is 2014 Jun;124(6):1290-5. susceptible to the segmentation approach implemented and manual segmentation had to be used. This is a minor issue but does not 2. Chambers KJ, Horstkotte KA, Shanley K, Lindsay RW. Evaluation of improvement in nasal obstruction following nasal valve correction hamper the methodology presented allowing – from a clinical in patients with a history of failed septoplasty. JAMA Facial Plastic perspective – sufficient alignment of simulated patient findings with Surgery. 2015 Sep 1;17(5):347-50. clinical data. Concerning the potential power of CFD simulation the presented results could be a promising step in the direction of 3. Prakash B. Patient satisfaction. Journal of Cutaneous and Aesthetic new tools for clinicians to predict anatomical areas and volumes to Surgery. 2010 Sep;3(3):151. surgically treat problems of nasal breathing. The results suggest that 4. Demirbas D, Cingi C, Çakli H, Kaya E. Use of rhinomanometry in a careful selection of the time-point of imaging is required so that common rhinologic disorders. Expert Review of Medical Devices. simulations actually can build on the “real” clinical situation and is 2011 Nov 1;8(6):769-77. not affected by changes to the nasal mucosa induced e.g., the topical 5. Januszewski M, Kostur M. Sailfish: A flexible multi-GPU application of local decongestants or the nasal cycle. implementation of the lattice Boltzmann method. Computer Physics Communications. 2014 Sep 1;185(9):2350-68. The 3D mesh was deformed randomly, approximating surgical changes and were subject to CFD simulations. 3D printed 6. Berger M, Pillei M, Mehrle A, Recheis W, Kral F, Kraxner M, et al. models with the changed geometry were validated against LDA Nasal cavity airflow: Comparing laser doppler anemometry and measurements. The root mean square error of the velocity near the computational fluid dynamic simulations. Respiratory Physiology & nasal valve of LDA measurements and LB simulations was 0.071. Neurobiology. 2021 Jan 1;283:103533. Changes in the geometry had similar effects on simulation and 7. Nakano H, Mishima K, Ueda Y, Matsushita A, Suga H, Miyawaki Y, measurement; the results found are in quite good agreement and et al. A new method for determining the optimal CT threshold for suggest that further investigations in this direction might present a extracting the upper airway. Dentomaxillofacial Radiology. 2013 fruitful route to pursue. Mar;42(3):26397438. 8. Dremel, “Dremel 3D20 - Operations manual.” 2015. Concerning the simulations of nasal stents, the data suggest that the resection volumes predicted are smaller when the stent is applied. 9. Alaxo GmbH: Neue Stent-Therapien bei Schnarchen, Schlafapnoe Consequently, the nasal stent “optimizes” the relevant available und behinderter Nasenatmung Alaxo GmbH: Neue Stent-Therapien intranasal volume for breathing since the proposed resection volume bei Schnarchen, Schlafapnoe und behinderter Nasenatmung. 2020. is minor. 10. Fedorov A, Beichel R, Kalpathy-Cramer J, Finet J, Fillion-Robin JC,

J Neurobiol Physiol 2021; 3(1):10-15. 14 Citation: Berger M, Pillei M, Freysinger W. The potential of computational fluid dynamics simulations of airflow in the nasal cavity. J Neurobiol Physiol 2021; 3(1):10-15.

Pujol S, et al. 3D Slicer as an image computing platform for the 12. Landau LD, Lifshitz EM. Quantum mechanics: non-relativistic Quantitative Imaging Network. Magnetic Resonance Imaging. 2012 theory. Elsevier; 2013 Oct 22. Nov 1;30(9):1323-41. 13. Giannarou S, Hacihaliloglu I. IJCARS-IPCAI 2020 special issue: 11th 11. Lorensen WE, Cline HE. Marching cubes: A high resolution 3D conference on information processing for computer-assisted surface construction algorithm. ACM Siggraph Computer Graphics. interventions-part 1. International Journal of Computer Assisted 1987 Aug 1;21(4):163-9. Radiology and Surgery. 2020 May 1:1.

J Neurobiol Physiol 2021; 3(1):10-15. 15