The inflow cannula is a crucial component of ventricular assist devices (VADs) with regard to cerebrovascular events.1–3 Malposition of the inflow cannula occurs in more than 50% of patients with left ventricular assist devices (LVADs).4 Malposition during surgery, postoperative pump migration, inflow obstruction, and right ventricular compression are major contributors to low flow and adverse events.1,4 Stroke in particular occurs at a high rate (8.7% incidence after device implantation) and is one of the leading causes of death in patients with continuous-flow devices, with 18.8% of deaths caused by neurologic dysfunction according to the INTERMACS 2017 third quarter report.5 Additionally, long-tipped cannulas can interfere with intraventricular flow. The EVAHEART 2 (Sun Medical Technology Research Corp., Nagano, Japan), with a novel inflow cannula and smaller pump, was developed to minimize the risk of inflow malposition and improve anatomical fitting and intraventricular flow dynamics to reduce the risk of postimplantation stroke.
The EVAHEART left ventricular assist system (LVAS) is a centrifugal pump meant to support patients in end-stage heart failure until a heart transplant becomes available. The pump contains an open-vane design to produce lower shear stresses and preserve von Willebrand factor.6 A unique impeller suspension mechanism, the “cool-seal” hydrodynamic system, was applied to the EVAHEART LVAS to levitate the impeller with continuously circulating sterile water through the driveline, pump body, and hydrodynamic bearing.
The original cannula design was a long-tipped cannula attached to a flexible conduit. However, the flexible conduit was prone to kinking over time, either through reverse remodeling of the heart or movement of the pump. Longer cannulas have been shown to be susceptible to occlusion, especially if they are inserted too deeply, with flow being restored once the cannula is moved.7 In addition, suction could occur if the cannula were to tilt toward the ventricular wall or septum as a result of malpositioning or remodeling.
Obstructed flow at the inflow by the myocardium can increase risk for adverse events such as pump thrombosis and ischemic stroke.1,8,9 Avoidance of right ventricular compression by the outflow graft is an additional key factor in preventing adverse events. To keep the right ventricular wall free of interference with the graft, the pump outflow should aim to the right of the sternal midline.1 To prevent the inflow cannula and pump body from pushing against the heart, it is important to make the pump pocket inferiorly deep and sufficiently lateral. According to Adamson et al.,1 pump pockets can shrink over time, and the pump has a tendency to be displaced superiorly and medially. Pump migration because of contraction of the pocket can change inflow cannula angulation, resulting in occlusion of the inflow tip.10 With the original inflow design of the EVAHEART LVAS, the pump pocket depth and size were quite large, compared with, for example, the HeartMate II. This led to a larger surgical invasion that may reduce surgical feasibility.
To design a novel inflow cannula for the EVAHEART LVAS that achieves maximal forgivingness to malposition during surgery, two major goals were identified:
- Optimal fitting of the inflow cannula and pump unit for a wide range of patients with the aim of reducing the risk of inflow obstruction, reducing pump pocket size, and avoiding right ventricle compression by the outflow graft.
- Optimal inflow cannula insertion length to reduce regions of high wall shear stress (WSS) and stagnation zones prone to thrombus formation in the ventricle.
The design optimization resulted in a short, rigid bend conduit with a double-cuff tipless (DCT) inflow cannula that significantly improves the anatomical fitting and washout of the left ventricle (LV). The DCT cannula is a low-profile design, allowing the inflow ostium to be flush against the endocardial plane at the LV apex (Figure 1). A rotational apex suture cuff draws in the surrounding myocardium to secure the fit and maintain the cannula’s flush position. In addition, the pump body was miniaturized (EVAHEART 2) to reduce the displacement volume of the system and allow pump implantation in smaller framed patients. Together, these factors minimize the risk of inflow cannula malposition and thrombus formation, allowing more forgivingness against malpositioning as a result of either the surgical procedure or remodeling of the heart and pump pocket.
Materials and Methods
To achieve the first goal in redesigning the inflow cannula, a virtual anatomy study using computed tomography (CT) scans of six representative patients with heart failure was performed. The novel design was iteratively adjusted to achieve ideal fitting and meet design requirements. Additionally, a fluid-structure-integration (FSI) model of the LV with a lumped parameter model of the entire cardiovascular system during VAD support was created and linked. Using this model, the hemodynamics of three inflow cannula insertion lengths for two patient-specific ventricles were calculated. Two support conditions, full support and partial support, were analyzed.
Contrast-enhanced high-resolution sectional image data acquired by CT scans were used for this study. The scans were acquired in end-diastolic phase, while the patient was in apnea. Six patient scans were selected, taking into account the quality of the scans, variety in sex, LV end-diastolic volume (LVEDV), and diversity of the medical history of the patients (Table 1). All data were investigated retrospectively and anonymously. The volumes and anatomy of the six scans differed sufficiently so that each inflow design could be tested for overall fit.
Patients 1 and 2 were potential candidates for a mechanical circulatory support device (one male and one female). Patients 3, 4, and 5 had aortic valve stenosis, with patient 5 having a very low LVEDV. A HeartMate II LVAD was implanted in patient 6. In total, the LVEDV ranged from 86 to 459 ml (Table 1).
Virtual anatomical models of the patients were created using the software Mimics and 3-matic (Materialise NV, Leuven, Belgium). The anatomical regions of interest were the left and right atrium, LV and right ventricle, pulmonary artery, aorta, myocardium, lungs, and thoracic cage. Segmentation of these regions was marked based on threshold values of grayscale in Hounsfield units followed by the use of automated standard image processing tools (e.g., static and dynamic region growing) and manual steps. Based on the final mask of each structure, a three-dimensional surface model was reconstructed. Finally, surface smoothing and wrapping was applied to fix inconsistencies in the model because of noise in the image.
Virtual Fitting and Design Optimization
The pump unit was superimposed on the physiologic structures in 3-matic by translation and rotation. Virtual fitting was performed for all six patient geometries and cannula configurations. The preferred position of the cannula was such that the cannula was parallel to the septum with the tip aiming toward the center of the LV or the mitral valve (Figure 2). Additionally, the pump outflow was aimed to the right of the sternal midline to avoid compression of the right ventricle.1
Three different designs of the inflow cannula were investigated and iteratively adjusted: the original long-tipped cannula with a flexible conduit; a hybrid, small-tipped cannula with flexible and rigid conduit components; and a DCT cannula with a rigid titanium conduit.
Each of the cannulas was first placed in the optimal position in parallel to the ventricular septum. If this brought the pump outside of the abdominal wall, the position was adjusted to fit. The resulting cannula position was evaluated for physical interference within the heart. Pump pocket depth was measured as the vertical distance between the apex of the heart and the lowest point of the pump body. Subsequently, the design of the cannulas was iteratively adjusted to achieve ideal fitting for all investigated patients. Parameters for design optimization were a) the bending angle of the rigid elbows, b) the offset between inflow and outflow, and c) the conduit length with the aim to reduce the risk of inflow obstruction, reduce pump pocket size, and avoid right ventricle compression by the outflow graft.
Numerical Simulation of Intraventricular Flow
Model structure and boundary conditions
To compute the intraventricular flow during VAD support, a one-way FSI model was chosen. The model structure is shown in Figure 3.
The LVs from patients 3 and 6 (Table 1) were used for the computational study to have a small and large ventricle for study. Two conditions, full VAD support (100% pump flow and 0% cardiac flow, called 100 VAD) and partial VAD support (70% pump flow and 30% cardiac flow, called 70 VAD), were investigated. The inflow cannula was positioned at a low initial orientation as well as at axial displacements of 1 and 2 cm. The methodological foundation can be found in the study by Liao et al.11
The fluid domain consisted of the LV geometry, generic inflow, and outflow tracts that represented the atrium and the aorta, and the inflow cannula. The structural domain consisted of the LV geometry with fixed supports at the boundaries of the inflow tract, the outflow tract, and the inflow cannula. During the simulation, the ventricular wall was deformed with the pressure pLV(t) and the wall deformation was translated into mesh displacement of the LV fluid domain. In the Computational Fluid Dynamics (CFD) simulations, the pressure profiles pLA(t) and pAo(t) were taken as pressure boundary conditions for the inlet and the outlet. During full support, the aorta was defined as a wall. An opening pressure boundary condition pVAD(t) (backflow allowed) was employed at the inflow cannula. The transient pressure profiles were computed with a lumped parameter network (LPN) of the entire cardiovascular system under VAD support conditions. Briefly, the LPN consisted of 14 different compartments characterized by pressure-flow and pressure-volume relationships with ordinary differential equations. The LPN was implemented and solved in MATLAB. The parameters were taken from Neidlin et al.12 and adjusted accordingly to represent the two investigated cases. The structure of the LPN, the parameter values, and the computed pressure profiles are found in Supplementary Material 1 (http://links.lww.com/ASAIO/A318).
To reduce computation time, heart valves were not considered in this study. The choice of a one-way FSI model over a two-way model was justified by the highly increased instabilities and remarkable increases in computational time in two-way FSI simulations.13,14
CFD mesh and simulation settings
The spatial discretization of the fluid domain was accomplished by an unstructured tetrahedral mesh with prismatic boundary layers using ANSYS ICEM (ANSYS, Inc., Canonsburg, PA). After a mesh independence study, the number of elements in models patients 3 and 6 was approximately 400,000 and 600,000, respectively. A time step of ts = 0.005 s was chosen for a total simulation time of 1.8 s, equivalent to two cardiac cycles. A Root Mean Square (RMS) convergence criteria of RMS = 1e-3 was set. The second cycle was chosen for result evaluation. Blood was modeled as a Newtonian fluid with a dynamic viscosity of 3.6 mPas and a density of 1056.4 kg m−3. Non-Newtonian behavior of blood was disregarded to improve the numerical cost.15 A mesh independence study, convergence plots, and a shear strain rate analysis are found in Supplementary Material 2 (http://links.lww.com/ASAIO/A319). All simulations were executed with ANSYS CFX (ANSYS, Inc.).
Finite element analysis mesh and simulation settings
The structural domain was discretized with shell elements of 5-mm thickness with an element number of 12,500 (patient 3) and 13,400 (patient 6). Because analyses of material stresses and strains were not in the scope of the study, the ventricular wall was modeled as an isotropic elastic material with a Poisson’s ratio of 0.3 and Young’s moduli ranging from 1.15 to 7 MPa to ensure deformation and stability for every simulation scenario. Structural wall deformation was transferred as fluid mesh displacement at every time step. Settings regarding the exact material parameters, the fluid mesh displacement, and the coupling settings are found in Supplementary Material 2 (http://links.lww.com/ASAIO/A319).
Results were evaluated for the three different implantation strategies regarding mean flows (pump and across aortic outlet), WSS, vortex ring formation, kinetic energy, and risk for thrombosis in the LV and inflow cannula. The kinetic energy (KE) of the LV was calculated via KE = 0.5 (u2 + v2 + w2), where u, v, and w are the respective velocity components of the flow.
Intraventricular vortices were identified with the Q criterion and visualized via iso-surfaces at a threshold of 475 s−2.16 Intraventricular vortex formation and kinetic energy of intraventricular blood flow can be connected to ventricular health as shown in several studies.17,18
Pump pocket depth was measured as the vertical distance between the apex of the heart and the lowest point of the pump body.
For the evaluation of thrombosis risk, many approaches exist without one perfect solution. These include blood residence time, pulsatility index, washout, and stagnation regions.11 A stagnation region approach was chosen that calculates the fluid volume in cells with strain rates <1.5 s−1 and v < 0.05 m/s, thus identifying recirculation regions.
Virtual Fitting and Design Optimization
Inflow cannula alignment
Implantation of the original configuration at the anterior wall was not possible for patients with LVEDV <400 ml because of interference with the ribs. After repositioning of the inflow cannula at the apex, the inflow tip faced toward the septum, which can lead to obstruction by the adjacent myocardium. By contrast, the hybrid and DCT configurations showed appropriate direction of the inflow cannula for all patients. For these configurations, the pump was placed at the preferred implantation site without abdominal wall interference (Figure 4).
Outflow graft and right ventricle compression
The avoidance of right ventricle compression was achieved with an inflow and outflow offset of 30° in all patients with all three configurations (Figure 5). This offset correlates with the findings of Adamson et al.,1 who investigated surgical principles for optimal HeartMate II positioning.
For patients 1, 3, and 5, the pump unit was positioned behind the sternal midline in the hybrid configuration. For patients with larger ventricles, this may be prevented with a deeper pump pocket (lower bending angle of the inflow conduit). However, for patients with a small LV (LVEDV <150 ml), this is not possible, and the pump unit may be significantly positioned behind the sternal midline. For the DCT, the pump unit was aimed toward the right of the sternal midline for all patient cases.
Pump pocket depth and size
Pump pocket depth for all three configurations is recorded in Table 2. The hybrid (mean: 117 mm) had a similar pump pocket depth to the original (mean: 126 mm). The DCT and smaller pump of the EVAHEART 2 had a significantly smaller pump pocket depth (mean: 98 mm) and volume, respectively, than did the other two configurations.
Inflow tip of tipless design
For patients 1, 2, and 5, with a wide LV and thin apical wall, there was no problem positioning the DCT. For patients with thicker myocardium, such as patients 3, 4, and 6, the DCT may not fully penetrate the ventricular wall without additional actions, which may result in interference with the myocardial trabeculae and consequently a higher risk of thrombus formation (Figure 6). In these cases, wedge resection and trimming of the trabeculae can be performed to prevent interference with the inflow ostium.
Hemodynamic Investigation of Inflow Cannula Insertion Length
Figure 7 shows the velocity profile on a coronal plane and intraventricular vortices (Q criterion = 475 s−2) for patient 6 during full support at time point t = 0.4 s, directly after the end of the A-wave (mid-diastole). Rotational movements of the flow in clockwise and counterclockwise directions are conceptualized by vector arrows.
Regions of high velocities are located around the vortices with maximum values of 0.33, 0.39, and 0.35 m·s−1 for insertion lengths of 0, 1, and 2 cm, respectively. The most significant difference between the three cannula insertion lengths was in the movement of the mitral vortex ring. When the cannula was inserted at 2 cm, the vortex collided with the cannula tip and was dissolved (dashed black frame). The same phenomenon was also recognizable in the partial support conditions. For the smaller ventricle (patient 3), no vortex dissolution due to cannula tip orientation was observed.
Systemic blood flow
The influence of cannula tip insertion length on systemic blood flow is illustrated in Figure 8A for the full support (Figure 8A, left) and half support (Figure 8A, right) cases. The cycle-averaged systemic flow of patient 6 was about 3 L/min for both cases. The cycle-averaged systemic flow of patient 3 was about 0.5 L/min for full support and 1.5 L/min for partial support. Cannula insertion length had minimal influence on hemodynamic performance.
Wall shear stress
To evaluate the risk of hemolysis from increased WSS, the WSS was evaluated in the area of the cannula tip during full support (Figure 8B) with the cannula at 0 cm leading to the lowest WSS. The first peak (t = 0.16 s) corresponds with the peak diastolic A-wave. The second maxima at 0.4 s (red), 0.45 s (green), and 0.55 s (blue) characterize when the mitral vortex ring collides with the cannula tip. To evaluate and compare all simulations, WSS was integrated over time and averaged over the entire cycle, to calculate the mean stress acting on the blood (Figure 8B, right). For patient 6, WSS were increased by 40% (0 cm in comparison to 2 cm). In patient 3, there were almost no changes in WSS.
Kinetic energy of the LV
For patient 6 on full support (Figure 8C, left), an insertion length of 1 or 2 cm reduced the peak kinetic energy by 50%, which was caused by the observed vortex ring dissolution. For a clearer comparison of the cannula impact, the total kinetic energy (area under curve) was computed and normalized (Figure 8C, right). A higher insertion length reduced the kinetic energy of the LV by 65% in patient 6 and by 15% in patient 3 (0 vs. 2 cm).
During VAD support, thrombus formation may occur around the cannula tip. Figure 8D (left) shows the node number prone to thrombus formation over one cardiac cycle. The insertion length of 2 cm greatly increased this risk. A comparison of all simulations showed that for patient 6, a cannula position differing from 0 cm was unfavorable (Figure 8D, right). Patient 3 showed no such behavior.
Inflow cannula design and positioning are crucial components in VADs with regard to adverse events. The virtual fitting described here allowed each design to be analyzed for best fit in a variety of LV morphologies, hemodynamic responses to cannula insertion depth, and forgiveness against malposition. The nature of this study is such that the methods can be applied when developing any new inflow cannula or LVAD pump. One drawback of a virtual fitting study, however, is the unavailable haptic feedback of necessarily compressed soft structures. Therefore, estimation of the resulting impact because of alterations of tissue is missing. A virtual fitting study does not replace cadaver studies or in vivo trials, but may reduce the number of such experiments needed, giving an economical benefit in creating new inflow designs. In addition, virtual fitting can be applied in a rather early stage of design development, thus reducing the number of design loops even before prototyping.
The smaller pump body of the EVAHEART 2 with the DCT cannula was unlikely to be obstructed even when the pump body had to be adjusted. With this new design, the heart remained in its natural anatomic position, thus ensuring long-term, unobstructed pump flow and potentially reducing the risk for adverse events in a wide range of patients. The implanted system also had a smaller pump pocket size than the other designs, making it a viable option for smaller patients and allowing for less surgical invasion. The 0-cm insertion length of the DCT cannula significantly reduced WSS levels, reduced thrombus risk, and preserved kinetic energy within the LV.
In comparison, the hybrid cannula was as forgiving against malposition in all LV sizes, but had the potential to bring the pump housing behind the sternal midline, especially in patients with a small LV. This could cause unnecessary compression of the right ventricle. In addition, this configuration had the largest pump pocket, making it unsuitable for smaller framed patients.
The original inflow cannula with the original EVAHEART pump housing was more likely to be obstructed with a higher possibility of touching the septum when the pump was adjusted, especially for patients with LVEDV <400 ml. In this case, the DCT cannula and EVAHEART 2 would provide the greatest forgiveness against malpositioning.
Analysis of the intraventricular hemodynamics revealed no influence of cannula insertion length on the pump or systemic blood flow. However, the cannula tip disturbed the formation of an intraventricular vortex that developed during cardiac diastole, moving from the base to the apex. The disturbance of this vortex also resulted in a decrease of total ventricular kinetic energy and an increase of WSS values at the cannula tip. Using vortex formation and kinetic energy as an index for cardiac health has been proposed in many studies17–19; thus, an interaction with the cannula tip might be unfavorable. Interestingly, these phenomena were only observed in the patient with a highly dilated ventricle (patient 6) and could not be confirmed in the patient with a small ventricle (patient 3). However, based on the classification seen in Table 1, patient 3 was not a VAD candidate from the beginning.
Validation of the computational analysis can be found in the study of May-Newman et al.,20 which used an in vitro setup to examine the influence of cannula insertion length on intraventricular hemodynamics. Although the ventricular model used in May-Newman et al.20 differed from that in our study, the major findings correspond with our conclusions. Reduction of kinetic energy and an interaction with the intraventricular vortex was observed with increasing cannula insertion length in May-Newman et al.20 Thromboembolic events were also investigated in May-Newman et al.,20 with the medium insertion length being chosen as the optimal condition. These findings did not correspond with ours; however, the method used to evaluate thrombus risk differed, thus reducing the comparability of the studies. Most strikingly, in the numerical study,11 various evaluation methods for thrombosis risk were implemented and an increased cannula insertion length was considered beneficial.
Such conflicting results emphasize the influence of ventricular geometry on study outcome and the need for better evaluation approaches of thromboembolic events during VAD support. Besides an improved thrombus risk evaluation method, several steps could be undertaken to improve the quality of the numerical model. First, the LPN was uncoupled from the CFD simulations, meaning that there is no interaction or feedback between these two models. Two-way coupling between a lumped parameter model and CFD simulations is feasible as shown in several studies.11,12,21 Second, there was no mitral valve in the current model. Including valve leaflets will strongly influence the formation of the intraventricular vortex with flow separation occurring at the tips of the mitral valve leaflets. An approach similar to Liao et al.11 that considered a parametric mitral valve model in an open and closed state might be useful to overcome the inherent challenges because of the high complexity of the mitral valve geometry and fully coupled fluid-structure interaction simulations.22
Despite these limitations, the presented model of LV blood flow with VAD support is able to show the interactions between intraventricular flow structures and cannula geometry. Because a rather generic design for the inflow cannula was used to investigate insertion length, the results of this study can be transferred to other VADs. In a further study, the hemodynamics (shear stress and flow separation) inside the inflow cannulas were compared by using numerical simulation. The DCT cannula showed the least flow dependence of possible risk factors. Regarding the shear stress, all three cannula designs were exposed to low levels of shear stress (average below 8 [Pa]).
The new DCT cannula and the smaller EVAHEART 2 show a better hemodynamic response among a wide variety of LV sizes compared with the other two cannula designs. This benefit combined with the improved forgiveness against malposition reduces the risk of adverse events in a wide range of patients.
1. Adamson RM, Mangi AA, Kormos RL, Farrar DJ, Dembitsky WP. Principles of HeartMate II implantation to avoid pump malposition and migration. J Card Surg 2015.30: 296–299.
2. Sacks J, Gonzalez-Stawinski GV, Hall S, et al. Utility of cardiac computed tomography for inflow cannula patency assessment and prediction of clinical outcome in patients with the HeartMate II left ventricular assist device
. Interact Cardiovasc Thorac Surg 2015.21: 590–593.
3. Taghavi S, Ward C, Jayarajan SN, Gaughan J, Wilson LM, Mangi AA. Surgical technique influences HeartMate II left ventricular assist device
thrombosis. Ann Thorac Surg 2013.96: 1259–1265.
4. Sorensen EN, Kon ZN, Feller ED, Pham SM, Griffith BP. Quantitative assessment of inflow malposition
in two continuous-flow left ventricular-assist devices. Ann Thorac Surg 2018.105: 1377–1383.
5. Parikh NS, Cool J, Karas MG, Boehme AK, Kamel H. Stroke risk and mortality in patients with ventricular assist devices. Stroke 2016.47: 2702–2706.
6. Bartoli CR, Kang J, Zhang D, et al. Left ventricular assist device
design reduces von Willebrand factor degradation: A comparative study between the HeartMate II and the EVAHEART left ventricular assist system. Ann Thorac Surg 2017.103: 1239–1244.
7. Crumpstone T, Martin TD, Yang JJ, Peng YG. Misplacement of LVAD inflow cannula leads to insufficient output and tissue hypoperfusion. J Artif Organs 2010.13: 225–227.
8. Kirklin JK, Naftel DC, Pagani FD, et al. Sixth INTERMACS annual report: A 10,000-patient database. J Heart Lung Transplant 2014.33: 555–564.
9. Starling RC, Moazami N, Silvestry SC, et al. Unexpected abrupt increase in left ventricular assist device
thrombosis. N Engl J Med 2014.370: 33–40.
10. Raman SV, Sahu A, Merchant AZ, Louis LB IV, Firstenberg MS, Sun B. Noninvasive assessment of left ventricular assist devices with cardiovascular computed tomography and impact on management. J Heart Lung Transplant 2010.29: 79–85.
11. Liao S, Neidlin M, Li Z, Simpson B, Gregory SD. Ventricular flow dynamics with varying LVAD inflow cannula lengths: In-silico evaluation in a multiscale model. J Biomech 2018.72: 106–115.
12. Neidlin M, Corsini C, Sonntag SJ, et al. Hemodynamic analysis of outflow grafting positions of a ventricular assist device using closed-loop multiscale CFD simulations: Preliminary results. J Biomech 2016.49: 2718–2725.
13. Neidlin M, Sonntag SJ, Schmitz-Rode T, Steinseifer U, Kaufmann TA. Investigation of hemodynamics during cardiopulmonary bypass: A multiscale multiphysics fluid-structure-interaction study. Med Eng Phys 2016.38: 380–390.
14. Sonntag SJ, Kaufmann TA, Büsen MR, et al. Simulation of a pulsatile total artificial heart: Development of a partitioned fluid structure interaction model. J Fluids Structures 2013.38: 187–204.
15. Chnafa C, Mendez S, Nicoud F. Image-based simulations show important flow fluctuations in a normal left ventricle: What could be the implications? Ann Biomed Eng 2016.44: 3346–3358.
16. Seo JH, Vedula V, Abraham T, et al. Effect of the mitral valve on diastolic flow patterns. Phys Fluids 2014.26: 121901.
17. Kanski M, Arvidsson PM, Töger J, et al. Left ventricular fluid kinetic energy time curves in heart failure from cardiovascular magnetic resonance 4D flow data. J Cardiovasc Magn Reson 2015.17: 111.
18. Pedrizzetti G, La Canna G, Alfieri O, Tonti G. The vortex–An early predictor of cardiovascular outcome? Nat Rev Cardiol 2014.11: 545–553.
19. Faludi R, Szulik M, D’hooge J, et al. Left ventricular flow patterns in healthy subjects and patients with prosthetic mitral valves: An in vivo
study using echocardiographic particle image velocimetry. J Thorac Cardiovasc Surg 2010.139: 1501–1510.
20. May-Newman K, Moon J, Ramesh V, et al. The effect of inflow cannula length on the intraventricular flow field: An in vitro
flow visualization study using the EVAHEART left ventricular assist device
. ASAIO J 2017.63: 592–603.
21. Migliavacca F, Balossino R, Pennati G, et al. Multiscale modelling in biofluidynamics: Application to reconstructive paediatric cardiac surgery. J Biomech 2006.39: 1010–1020.
22. Wu W, Pott D, Mazza B, et al. Fluid-structure interaction model of a percutaneous aortic valve: Comparison with an in vitro
test and feasibility study in a patient-specific case. Ann Biomed Eng 2016.44: 590–603.