A finite strain nonlinear human mitral valve model with fluid-structure interaction

A computational human mitral valve (MV) model under physiological pressure loading is developed using a hybrid finite element immersed boundary method, which incorporates experimentally-based constitutive laws in a three-dimensional fluid-structure interaction framework. A transversely isotropic material constitutive model is used to characterize the mechanical behaviour of the MV tissue based on recent mechanical tests of healthy human mitral leaflets. Our results show good agreement, in terms of the flow rate and the closing and opening configurations, with measurements from in vivo magnetic resonance images. The stresses in the anterior leaflet are found to be higher than those in the posterior leaflet and are concentrated around the annulus trigons and the belly of the leaflet. The results also show that the chordae play an important role in providing a secondary orifice for the flow when the valve opens. Although there are some discrepancies to be overcome in future work, our simulations show that the developed computational model is promising in mimicking the in vivo MV dynamics and providing important information that are not obtainable by in vivo measurements. © 2014 The Authors. International Journal for Numerical Methods in Biomedical Engineering published by John Wiley & Sons Ltd.


INTRODUCTION
Dysfunction of the mitral valve (MV) causes significant morbidity and mortality and remains a major medical problem worldwide. Understanding the biomechanics of human MVs can lead to the development of new therapies and treatment strategies. Changes in MV geometry could perturb the stress patterns and affect repair durability [1,2]. However, measuring detailed stress patterns in MVs is extremely challenging in vitro and nearly impossible in vivo. Computational models can fill 1599 important for accurately predicting the transvalvular flow patterns. Most of these studies used the commercial package LS-DYNA (Livermore Software Technology Corporation, Livermore, CA) to implement FSI . We use open source implementations of IB methods for FSI [36,[44][45][46] to simulate MV dynamics. Over the last few years, we have applied the IB method to develop three-dimensional FSI models of a polyurethane prosthetic MV. In a series of studies, we considered the effects of the chordae in the MV [45,46], of the pressure loading condition [35], and of the left ventricular motion [47], which was prescribed using high resolution MRI from a normal human ventricle. We also found that, despite the very thin MV leaflets, their bending rigidity is highly relevant for the effective closing of the MV [48]. The bending effect of the MV is also confirmed by a recent study of human MV [44] based on in vivo MRI, which shows that patient-specific MV geometry has a significant influence on the simulation results. One major limitation of these studies is that the MV is modelled with discrete isotropic 'elastic fibres', which are not readily able to model the anisotropic hyperelastic mechanical behaviour of the MV leaflets as FE MV models [7,11].
The aim of this study is to overcome the structural simplifications of our previous MV models by developing a combined IB and FE MV model. FE-based structural models for the IB method have been introduced recently [49][50][51][52][53]. One of the schemes, a hybrid finite difference-FE IB method (IB/FE) [54], has been carefully verified against commercial package ABAQUS [55], and the results show that the IB/FE model is capable of predicting quantitatively accurate stress/strain distributions for realistic biomechanical models of the heart. In this study, our MV model is again constructed from a human MRI study, but here, the IB/FE approach will enable us to model the MV tissue behaviour using a transversely isotropic model together with the fully dynamic three-dimensional FSI. This is a major step forward from our previous work [44].

The IB/FE method
The IB/FE approach [54] uses a Lagrangian FE description of the immersed structure along with an Eulerian finite difference description of viscous incompressible fluid, which is modelled using the incompressible Navier-Stokes equations. Let R 3 denote the physical domain occupied by the fluid-structure system, and let U R 3 denote the reference coordinate system attached to the MV. Let x D .x 1 ; x 2 ; x 3 / 2 denote fixed physical coordinates, let X D .X 1 ; X 2 ; X 3 / 2 U denote material coordinates attached to the structure, and let .X; t/ 2 denote the physical position of material point X at time t , so that .U; t / D s .t / is the physical region occupied by the structure, and the physical region occupied by the fluid at time t is f .t / D n s . Let N.X/ denote the exterior unit normal to X 2 @U in the reference configuration. The IB/FE formulation of the equations is Â @u @t .
f.x; t/ D Z U r X P e .X; t/ ı.x .X; t// dX Z @U P e .X; t/ N.X/ ı.x .X; t// dA.X/; (4) in which is the mass density, is the viscosity, u.x; t/ is the Eulerian velocity field of the fluid-structure system, p.x; t/ is the Eulerian pressure field, f.x; t/ is the Eulerian elastic force density, and ı.x/ D ı.x 1 / ı.x 2 / ı.x 3 / is the three-dimensional Dirac delta function. We use the first Piola-Kirchhoff stress tensor P e to describe the stresses generated by the immersed structure. P e is determined from the passive hyperelastic properties of the leaflets by a strain-energy functional W through P e D @W @F , in which F D @ =@X is the deformation gradient associated with the structural deformation. Thus, the total Cauchy stress of the coupled fluid-structure system is f is the fluid-like stress tensor defined as in which p D p.x; t/ is the hydrostatic pressure and e is the elastic stress tensor that is related to P e by P e D J e F T , with J D det.F / Although we refer to e and P e as the structural stress, we remark that they are not the total stresses of the immersed structure but only account for the stresses associated with the hyperelastic material response. The total stress of the immersed structure is D pI C OEru C .ru/ T C e ; see Eq. (5). Eq. (3) specifies that the velocity of the immersed structure is derived from the Eulerian velocity field, which implies @ @t .X; t/ D u. .X; t/; t/: Notice that Eq.
In the numerical implementation, the Eulerian equations of motion are discretized by a finite difference method, and the Lagrangian equations are discretized with an FE method. Therefore, an equivalent weak formulation of the definition of f in (4) is introduced as Z U F.X; t/ V.X/ dX D Z U P e .X; t/ W r X V.X/ dX; (9) in which F.X; t/ is the Lagrangian elastic force density and V.X/ is an arbitrary Lagrangian test function that is not assumed to vanish on @U . Although Eqs. (4) and (8)- (9) are equivalent in the continuous setting, they lead to different numerical schemes, and Eqs. (8)-(9) allow F.X; t/ to be determined via a standard total Lagrangian FE scheme. Further details are provided by [54]. The simulations described herein employ the open-source IBAMR software framework, which provides an adaptive and distributed-memory parallel implementation of the IB method and an infrastructure for developing FSI models that use the IB method. IBAMR leverages functionality provided by other freely available software libraries, including SAMRAI (https://computation-rnd. llnl.gov/SAMRAI), PETSc (http://www.mcs.anl.gov/petsc) and hypre (https://computation-rnd.llnl. gov/linear_solvers).

MV geometrical model
An MRI study was performed on a healthy 28-year-old male volunteer using a 3-Tesla MRI system (Verio, Siemens, Germany). The study was approved by the local NHS Research Ethics Committee, and written informed consent was obtained before the scan. Twelve planes along the left ventricular outflow tract (LVOT) view were imaged to cover the entire MV. Typical parameters are: slice thickness, 3 mm with 0 mm gap; matrix size, 432 572; in-plane pixel size, 0:7 0:7 mm 2 ; and frame rate, 25 per cardiac cycle. Short-axis and long-axis views of the ventricle were also acquired to provide additional geometrical information, such as the heads of papillary muscles.
The MV was segmented in the middle of diastole when it is opened by using an in-house MATLAB (The MathWorks Inc., Natick, USA) code. The segmentation method is similar to that in the previous work [44]. Detailed steps are described in the Appendix. The MV leaflet are meshed with 154,000 tetrahedral elements with at least two elements across the leaflet thickness.

Material models
The leaflets of the MV are modeled as an incompressible fibre-reinforced material, in which the strain-energy is a functional of invariants of the right Cauchy-Green deformation tensor C D F T F that takes the form in which I 1 D trace.C/ is the first invariant of the right Cauchy-Green deformation tensor and I f D f 0 .Cf 0 / is the square of stretch in the fibre direction, with f 0 denoting the unit tangent to the fibre direction field in the reference state. The modified invariant I ? f is defined in terms of I f by This ensures that the load-bearing collagen fibres embedded in the MV leaflets only bear the loads in extension but not compression. The material parameters C 1 , a f , and b f were fit to equal-biaxial in vitro tests on a healthy human MV carried out by [11] as shown in Table I. Following our previous study [55], we found that it is useful to employ a modified structural stress tensor P e defined as in which I 3 D det.C/ D J 2 andˇs D 500 kPa. The pressure-like term C 1 F T ensures that for F D I; P e D 0, which reduces the magnitude of the pressure discontinuity in the Eulerian pressure field and thereby the magnitude of the spurious volume loss at fluid-solid interface. The termˇs log.I 3 /F T is a penalty term to reinforce the incompressibility constraint in the Lagrangian form, because even though incompressibility is enforced in the Eulerian equations, the numerical interpolation of the Eulerian velocity to the solid region may not always yield a divergence-free discrete Lagrangian velocity field. This additional constraint has been shown to yield more accurate stresses [55]. Although linear tetrahedral elements will yield volumetric locking for sufficiently large values of s , comparisons to the same MV model withˇs D 0 verified that for the value ofˇs used in this study, our model does not experience volumetric locking because the two models have comparable deformations under the same static loading.
The chordae tendineae are assumed to be isotropic and modelled as a neo-Hookean material: in which C is spatially homogeneous but takes the value of 9.0 MPa in systole and 0.54 MPa in diastole. These values are based from the measurements of human MV chordae [56]. Similarly, the structural tensor P e in the chordae tendineae is defined as

Boundary conditions and implementation
In the simulations, we set D 1:0 g/ml and D 4 cP, and we fix the MV annulus to a housing disc that is mounted in a rigid circular tube immersed in a 10 cm 10 cm 16 cm fluid box ( Figure 1). The fluid box is discretized with spacings x 1 D x 2 D 0:125 cm; x 3 D 0:1 cm, corresponding to a regular 80 80 160 Cartesian grid. In the numerical scheme, the singular delta function kernel appearing in Eqs. (3) and (8) is replaced by a standard four-point regularized version of the delta function [36]. The integral transforms appearing in the interaction equations are approximated using dynamically generated Gaussian quadrature rules that ensure a density of at least two quadrature points per Cartesian mesh width. After computational experimentation, a time step size of 2.5e-5 s   (a) is chosen in the explicit time stepping scheme. Further details of the spatial discretizations and time stepping scheme are provided by [54]. We specify the pressure differences between the inlet and outlet of the tube. Because subjectspecific transvalvular pressure data are not available, we instead use a typical human physiological pressure profile rescaled to the cuff pressure of the subject, so that the peak systolic pressure is 150 mmHg § ; see Figure 2. We do not model the left ventricle, so only the rapid filling phase of the diastole and the systolic phase are modelled here. This is because in the slow diastolic filling § Note that this subject's blood pressure pressure is slightly higher than average.  and atrial contraction, the transvalvular flow results mostly from the passive deformation of the left ventricle. Without a model of the left ventricle, slow diastolic filling and atrial contraction cannot be modelled with the proper boundary conditions. This is different from rapid filling, in which the transvalvular flow is due to the suction caused by the left ventricular relaxation. In systole, the boundary conditions can also be easily approximated with the systolic pressure and the papillary muscle movement, as done in other studies [48]. The rapid diastolic filling lasts for around 0.2 ms, during which time 80% of transvalvular flow occurs [57]. The pressure profile for the part of the cardiac cycle we modelled is shown in Figure 2. Zero-pressure boundary conditions are applied along the remainder of the domain boundaries, similar to our previous study [44].
A grid convergence study is performed with 80 80 160; 96 96 198, and 112 112 224 Cartesian grids. Table II shows the average and the maximum displacement of the MV model when fully opened and loaded. Notice that only small differences in the computed values are observed for 80 80 160 when compared with two denser Cartesian grids. For computational efficiency, we use the 80 80 160 grid for all subsequent simulations.

RESULTS
The fluid pressure fields from the IB/FE MV model are shown in Figure 3 at the three time instants indicated in Figure 2: when the valve is fully opened (t D 0:1 s); just closed (t D 0:22 s); and fully loaded at the peak systolic pressure (t D 0:35 s). The valve opens at a driving pressure gradient of approximately 10 mmHg and withstands a significant physiological transvalvular pressure gradient when closed. The complete closure of the leaflets occurs at a transvalvular pressure gradient of around 80 mmHg and then undergoes minor further deformation before reaching the peak transvalvular pressure difference (150 mmHg), as shown in Figure 3(b) and (c). Notice that our MV model has small gaps near the commissure area even in the fully closed state, although these are very small and there is no flow leakage in the IB/FE simulations. The corresponding velocity fields are shown in Figure 4 and indicate a strong jet flow towards the outlet (the left ventricular side) when the MV opens. As the MV closes, there is a closing regurgitation across the MV, as shown in Figure 4(b), which is responsible for the first heart sound. Figure 4(c) shows the velocity pattern when the transvalvular pressure difference peaks. An interesting view is provided by the streamlines when the MV is fully opened, as shown in Figure 5. Here, the flow is first channelled by the two leaflets, then the chordae act as the second orifice [58] and split the jet flow into three paths, with the main stream moving towards the LV apex and side streams flowing towards the walls.
The deformed MV leaflets are compared with the corresponding MRI in LVOT view in Figure 6. Qualitatively, the opening and closing configurations of the simulated MV show good agreement with the MRI measurements. However, there is evidence of some discrepancy, particularly in the anterior leaflet when fully closed: the modelled MV bulges into the left atrium, in contrast with that of the corresponding MRI (Figure 6(c)). This is presumably because of simplifications in the chordal model, or the lack of strut chordae.
The flow rate through the valve is shown in Figure 7 along with the measured flow rate from phase contrast MRI, as well as the flow rate obtained from a fibre-based IB MV model [44]. When the MV is open, the flow rates from both the IB and IB/FE MV models are comparable with the measured value, with slightly lower peak values. However, the IB/FE model seems to predict a slightly higher regurgitation closing flow (10.8 mL vs. 9.4 mL), followed by much smaller oscillations compared with the previous IB MV model.
The stress and strain distributions along the fibre directions are shown in Figure 8. When the MV is fully open, most of the two leaflets are stretched along the fibre direction, and the stress Figure 7. Flow rate comparison between the predictions of the IB/FE model, a previous IB model [44], and the flow measurements obtained from a phase contrast cine magnetic resonance imaging study. Note that only the measured flow rate in rapid filling phase of diastole is plotted, which lasts for 0.2 s.  Table III. Average regional stress and strain along the fibre direction on the three local regions defined in Figure 9 when the mitral valve is fully opened, just closed, and fully loaded with peak systolic pressure. level is lower (Figure 8(a) and (d)). When the MV is just closed, high stress concentrations occur in the fibrous trigones of the anterior leaflet and along the valvular ring. Towards the commissures, there are compressive stresses, particularly in the neighbourhood of the wrinkles. When approaching the highest systolic pressure, the strain and stress patterns (Figure 8(c) and (f)) are similar as the distributions when the MV is just closed (Figure 8(b), (e)), but the high stress area increases, which is most visible in the belly and annular ring regions of the anterior leaflet and higher than the posterior leaflet.
We sampled three local regions in the anterior leaflet for detailed analysis of stress and strain. These include two trigones and one belly region, as shown in Figure 9. The average stress and strain of these regions are summarized in Table III. We can see that when the MV is fully open, the fibres of the belly region are stretched, as indicated by the positive strain, and the stress is higher compared with the trigone regions. Immediately after the MV has closed, the stresses increase in all regions; however, at this point, the two trigones experience higher stress levels compared with the belly region. This pattern remains the same when the MV experiences systolic pressure loads, although the stresses are nearly doubled. Table IV summarizes the average chordae tendineae tension when the MV is fully loaded with the maximum systolic pressure of 150 mmHg. The average force for each chordae associating with the anterior leaflet is 0.68˙0.31 N, which is substantially higher than that of the chordae of the posterior leaflet. The tensions in the two papillary muscles are 3.0 N and 3.34 N, respectively. The tension experienced by the chordae and the papillary muscles are comparable with values from other studies [7,60].

DISCUSSION
In this paper, we presented a new IB/FE model for a human MV using in vivo MRI-based anatomy and fibre-reinforced hyperelastic laws. This approach allows us to predict the stress and strain patterns in the MV leaflets with experimentally determined constitutive parameters, which is not possible using our previous IB models [44], as well as to include FSI dynamics. The model is validated by comparing the computed opening and closing configurations and flow rates to clinical measurements from the same volunteer from whom the MV model is constructed.
IB methods have been widely used to simulate the dynamics of the heart and its valves [35,44,48,59,[61][62][63]. Most earlier IB MV models used discrete collections of elastic fibres to represent the real MV structure [44]. Although these 'explicit fibre' models are well suited for describing extremely anisotropic materials, it is challenging to link such descriptions to experimental data and to use hyperelastic constitutive laws designed for fibre-reinforced soft tissue mechanics. Recent extensions of the IB method are able to incorporate finite strain elasticity models for the solid region [49][50][51][52][53], including that used herein. Our results show that this combined IB/FE approach is able to generate a three-dimensional FSI model that can simulate in vivo MV dynamics during diastole and systole.
Compared to our previous work [44], significant improvements are incorporated in the current IB/FE MV model. Namely: (1) the MV leaflets are described with a continuum mechanics framework; (2) the annulus ring is reconstructed from clinical images, not an assumed planar elliptic shape; (3) transversely isotropic constitutive laws are used for the MV leaflets, not the isotropic linear material model; and (4) detailed stress and strain fields are determined, which are not available in our previous IB MV model [44].
Using cine MRI, MV leaflet delineation is most clearly defined in diastole when the transvalvular pressure difference is minimal because the influences from blood flow are minimized [64] and the MV is fully open. By sewing miniature markers onto the MV leaflets of 57 sheep, Rausch et al. [65] calculated in vivo strain in the anterior leaflets using reference states corresponding to minimum left ventricular pressure, end-diastole, and the beginning of rapid diastolic filling, and they found that the strain and strain rates are almost insensitive to the choice of the reference configurations. With this in mind, we reconstruct the MV structure in mid-diastole, and because the transvalvular pressure difference is minimal, this is also close to the zero loading state. We also keep the mitral annulus fixed in space, but allow the papillary muscle to move towards the leaflets during systole with a maximum displacement measured from the corresponding cine MRI. Doing this allows the leaflets to close when subjected to a physiological pressure load, as we found previously [46,47].
From Figure 8(a), (b), and (c), it is clear that the strains along the fibre directions are tensile for most of the regions in the anterior leaflet during the cardiac cycle, but some regions that are compressed during closing. Similar results have been reported previously [27,41]. Our model prediction of the circumferential strain range seems to agree well with the measurements from in vivo measurements. For example, Rausch et al. [65] measured the peak in vivo circumferential strain (along the fibre directions in this study) in the belly region of the anterior leaflet, which is 3:5˙3:6%. By defining a circular region with a 0.5 cm radius in the belly region of the anterior leaflets, the average strain along fibre directions from our IB/FE model is 3:6˙4:0%, which is comparable with Rausch's results and also lies in the range reported by Sacks et al. [66] from in vivo studies (2:5 3:3%).
On the other hand, mitral leaflet strain estimated from some in vitro measurements seem to be higher than our predictions. Jimenez et al. [67] measured strain in the (porcine) anterior leaflet centre using a left heart simulator, and they found that their peak circumferential strain is 11˙4.9% in the normal annular configuration. Similar peak circumferential strain (around 10%) in the leaflet center was also reported in the in vitro studies by Sacks et al. [68] and He et al. [69,70], although the range of the peak circumferential strain can be as low as 2% [70] (within one standard deviation). The difference in strain between the in vivo and in vitro measurements was also noted by Rausch et al. [65]. It is not clear if the difference is due to the general difference between the porcine (used in the in vitro studies), ovine (used in the in vivo studies), and human valves (used in our study), or by the incapability of in vitro models to accurately reproduce the mechanical and hemodynamic environment (in terms of the material property, boundary condition and the initial geometry) of the MV. We also note again that the peak systolic pressure of 150 mmHg used in the simulation is slightly higher than a typical value (100-140 mmHg).
The peak average stress along the fibre direction in the belly region is 115 kPa and is 169 kPa and 134 kPa for the two fibrous trigons (Table III) These predicted stress levels are of similar order to results of other published MV models [5,41]. Jimenez et al. [60] also measured different types of chordae systolic tension and the forces experienced by the papillary muscles in their in vitro experiments. They reported the peak systolic tension of 0.95˙0.35 N in the anterior strut chordae and of 0.35˙0.16 N in the anterior marginal chordae. In our model, the average tension in the anterior chordae is 0.68˙0.31 N, which is comparable with the range reported by Jimenez [60] and the model results by Wang et al. [11]. The maximum papillary muscle forces are found to be 4.3 N and 4.6 N from Jimenez's experiments [60], and 4.51 N and 5.17 N from Wang's model [11]. These are also comparable with our predications of 3.0 N and 3.34 N.
Although our results are in good agreement with MRI observations and measured flow rates, some model limitations are worth mentioning. For example, the modelled anterior leaflet appears to bulge into the left atrium, in contrast with that of the cine MRI. Also, there are small gaps in the commissure areas when the valve is fully closed. This is due to the difficulties of reconstructing the subvalvular apparatus accurately from cine MRI, especially for the commissure areas and chordae tendineae. We have to build these relying on anatomical descriptions widely adopted in other studies [9,10,30,33,44]. More accurate geometry reconstruction may be possible using computed tomography (CT) images. For example, Wang and Sun [11] constructed MV models with chordal origins and insertion points by using multislices CT images. However, CT scans may not be used with healthy human subjects because of the radiation risk. Real time three-dimensional echocardiography could be another promising way to obtain detailed MV structure either noninvasively or with limited invasiveness [30], although the reproducibility is not as good as with CT and MRI. Current in vivo MRI is not able to quantify the fibre directions in the moving MV leaflets, and therefore their distributions are usually modelled using 'rule-based' methods [7,30,33,41,44], following experimental observations [6]. Recently, Lee et al. [71] developed a microanatomical MV models from in vitro experimental measurements by mapping the measured collagen fibre architecture using the small light scattering techniques to the MV models from the same ovine MV leaflets. Non-invasive methods are needed to quantify the in vivo fibre architecture.
Moreover, we have only used simplified primary chordae inserting into the leaflets from the free edges, following our previous work [44][45][46]. In particular, we have not reconstructed the marginal, basal and strut chordae separately, which have different mechanical properties. This may be partially responsible for the MV bulging towards the left atrium. Chordae are bundles of collagen fibres, and some groups modelled these using anisotropic hyperelastic constitutive laws [7,11]. We model the chordae as isotropic neo-Hookean material because in diastole, the chordae are usually compressed when MV opens and behave like an isotropic material. In systole, the chordae are stretched and a much higher stress is generated because of the stretched collagen fibres. This is modelled using a much higher chordae stiffness in the systole. Overall, this is a simplified model of the chordae structure that has reproduced the measured in vivo dynamics of MV. With new developments in clinical imaging it may be possible to model more detailed patient-specific chordae structure and its effect on the MV behaviour in the future.
Finally, our initial configuration is partially open, and it is in general very difficult to fully close an opened complex MV configuration that is not initially driven from a fully closed state. Inaccurate geometric details in the commissure areas may be partially responsible; another reason may be that the flows around the MV are not fully physiological due to the absence of the left ventricle.

CONCLUSION
In this study, we have developed a new human MV model using a hybrid IB/FE approach, which combines FSI simulations with cine MRI and a transversely anisotropic, hyperelastic constitutive model. This model is a significant enhancement from our previous work because it provides dynamic stress distributions, which are found to be concentrated around the annulus trigons and the belly of the anterior leaflet. The model results agree well with the opening and closing leaflet configurations, and with flow rates, estimated from MRI measurements. Although there are still some discrepancies between the model predictions and in vivo observation, with improved imaging techniques and further work, it is possible to develop more realistic MV models that could be applied to study MV diseases.

A. Detailed segmentation of the human MV model
Leaflet segmentation: the anterior and posterior profiles were manually segmented on cine images of LVOT views ( Figure A.1(a)). An uniform leaflet thickness of 1 mm was assumed, as shown in Figure A.1(c). Chordae tendinae: A total of 16 evenly distributed primary chordae tendinae were defined based on anatomical descriptions, with 10 associated with the posterior leaflet and six with the anterior leaflet. In our model, the chordae attach to the free margin of the leaflets and run through the leaflets to the annulus ring. These structures model the function of the chordae and we have not separated the contributions from the marginal chordae, strut chordae, and secondary chordae due to the lack of information from the MR images. Similar chordae structure was used in our previous study [44]. Each of the chordae is connected to one of the two papillary muscle heads as shown in Figure A.2(a), and each is assumed to have a uniform crosssectional area of 0.16 mm 2 , with a thickness of 0.4 mm and average length of 32˙6 mm. The papillary heads are allowed to move towards the MV leaflets during systole with displacements measured from the cine MR images, which permits the leaflets to close when subjected to a high systolic pressure. Collagen fibre architecture: The embedded collagen fibres were defined along the circumferential direction (parallel to the annulus ring) [3] for both the anterior and the posterior leaflets, as shown in Figure A.2(c) and Figure A.2(d).