ARTICLE IN PRESS

Journal of Biomechanics 41 (2008) 25392550 www.elsevier.com/locate/jbiomech www.JBiomech.com

Numerical simulation of the dynamics of a bileaet prosthetic heart valve using a uidstructure interaction approach
Matteo Nobilia, Umberto Morbiduccib, Raffaele Ponzinic, Costantino Del Gaudiod, Antonio Balduccid, Mauro Grigionid, Franco Maria Montevecchib, Alberto Redaellia,
Department of Bioengineering, Politecnico di Milano, P.za Leonardo da Vinci, 32, 20133 Milano, Italy b Department of Mechanics, Politecnico di Torino, Turin, Italy c CILEA, Interuniversity Consortium, Milan, Italy d `, Italy Cardiovascular Bioengineering Unit, Technology and Health Department, Istituto Superiore di Sanita Accepted 6 May 2008
a

Abstract The main purpose of this study is to reproduce in silico the dynamics of a bileaet mechanical heart valve (MHV; St Jude Hemodynamic Plus, 27 mm characteristic size) by means of a fully implicit uidstructure interaction (FSI) method, and experimentally validate the results using an ultrafast cinematographic technique. The computational model was constructed to realistically reproduce the boundary condition (72 beats per minute (bpm), cardiac output 4.5 l/min) and the geometry of the experimental setup, including the valve housing and the hinge conguration. The simulation was carried out coupling a commercial computational uid dynamics (CFD) package based on nite-volume method with user-dened code for solving the structural domain, and exploiting the parallel performance of the whole numerical setup. Outputs are leaets excursion from opening to closure and the uid dynamics through the valve. Results put in evidence a favorable comparison between the computed and the experimental data: the model captures the main features of the leaet motion during the systole. The use of parallel computing drastically limited the computational costs, showing a linear scaling on 16 processors (despite the massive use of user-dened subroutines to manage the FSI process). The favorable agreement obtained between in vitro and in silico results of the leaet displacements conrms the consistency of the numerical method used, and candidates the application of FSI models to become a major tool to optimize the MHV design and eventually provides useful information to surgeons. r 2008 Elsevier Ltd. All rights reserved.
Keywords: Prosthetic heart valve; Computer modelling; Finite-volume method; Dynamic analysis; Cardiovascular biouid mechanics; Fluidstructure interaction; Parallel processing

1. Introduction Articial heart valves are designed to replicate the function of the natural valves of human heart. Since their introduction about 50 years ago (DeWall et al., 2000; Gott et al., 2003), the study of their uid mechanics became an important research eld in bioengineering. While the design of the early prototypes was driven by a quantitative analysis of the overall ow eld downstream of the valve,
Corresponding author. Tel.: +39 2 239 93 375; fax: +39 2 239 93 360.

E-mail address: alberto.redaelli@polimi.it (A. Redaelli). 0021-9290/$ - see front matter r 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbiomech.2008.05.004

more recent studies focused their attention on the local uid dynamics occurring near the valve housing (Leo et al., 2002, 2006; Kelly, 2002) and in proximity of the leaets, which are areas commonly associated with thrombosis (Yin et al., 2004). The geometrical complexity of prosthetic mechanical heart valves (MHVs), joined with the nonlinear, dynamic nature of the interplay between the pulsatile owing blood, the complex ventricular and aortic anatomies, and the valve mobile components, renders the study of ows through MHVs a challenging issue for even the most advanced computational and experimental techniques

ARTICLE IN PRESS
2540 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550

(Yoganathan et al., 2004, 2005; Grigioni et al., 2004). In particular, limitations in performing fully three-dimensional quantitative observations of the local ow eld by any of the currently available laboratory techniques like particle image velocimetry and laser Doppler anemometry, that are present even in controlled, relatively simplied in vitro conditions, have suggested the numerical method as an additional instrument for understanding the dynamics of owing blood. In this context the use of computational uid dynamics (CFD) has gained relevance for heart valve assessment (Alemu and Bluestein, 2007; Dumont et al., 2007; Dasi et al., 2007; Guivier et al., 2006; Bluestein et al., 2000). Indeed, as computational resources become more powerful, and data handling algorithms become more sophisticated, numerical analysis has became an increasingly important part of biomechanics research, allowing the modelling of even more realistic geometries, and achievement of high spacetime resolution. Two different approaches are currently used for computational investigations. The rst approach neglects the interplay between leaets and blood, focusing on the peak and near peak systole phase of the cardiac cycle. In this research framework, Grigioni et al. (2005a) performed a numerical simulation on a model of mechanical valve placed in a physiological aortic root shaped model, furnishing indications about the role of Valsava sinus on the uid dynamic downstream of the valve at mid systole. Yokoyama et al. (2006) demonstrated that curved leaet conguration generates a Venturi effect that is able to reduce the boundary layer separation zones and stabilize the leaets position. Ge et al. (2003) provided an insight into the complex hemodynamics of a bileaet MHV at the maximum Reynolds number, putting in evidence that high grid densities are mandatory, in order to fully characterize local aspects of the ow eld downstream of the prosthetic valve. Alternatively, the investigation of bloodleaet interplay through the application of uid structure interaction (FSI) models moves into the direction of providing more indepth insight into the MHVs hemodynamics, including the valve opening and closure assessment. More in depth, the rationale that moves research is that development of numerical methods/strategies to predict even more realistic biouidstructure interaction phenomena related to implanted devices could:

 

provide useful responses to surgeons (choice of the best implantation strategy); reduce costs to manufacturers (physical modelling can be a very effective and accurate method of research and testing, but it can be very expensive and time consuming due to the numerous tests that must be conducted); lead to a better comprehension of those physiopathological processes for which a multiphysics, multiscale approach is needed (and for which computer modelling is the strategy of election).

A hybrid approach is the one proposed by Lai et al. (2002), who studied the closure phase of a prosthetic heart valve using a prescribed leaet motion in a simplied ow domain. Very recently, direct numerical simulation (DNS) of pulsatile ow through a bileaet MHV mounted in an idealized axisymmetric aorta geometry (with a sudden expansion modelling the aortic sinus region) was carried out by prescribing leaet motion from the experimental data (Dasi et al., 2007). However a prescribed kinematics avoids the complex phenomena of uidleaet interaction required for an accurate simulation of the valve function. Therefore FSI algorithms have been developed to predict the motion of the leaets, which is the direct consequence of the momentum applied by the uid on them (Penrose and Staples, 2002; Vierendeels et al., 2005; Nobili et al., 2007). FSI studies generally focus on a specic phase of the valve dynamics and introduce geometrical simplications because of the complexity of the problem and the computational costs. Redaelli et al. (2004) analyzed the opening phase of a bileaet heart valve under low ow rates and validated the leaet motion experimentally. Cheng et al. (2004) presented a three-dimensional unsteady ow analysis past a bileaet valve prosthesis in the mitral position during the closing phase, incorporating the leaet motion in their model. The whole cardiac cycle was simulated by Dumont et al. (2005) in laminar condition, applying a symmetry assumption to the ow. Lately Pelliccioni et al. (2007) presented a bidimensional study of MHV dynamics in laminar ow condition by applying the latticeBoltzmann method. The aim of this study is the investigation, in silico, of the bileaet mechanical valve dynamics during the whole systolic phase. For this purpose, a fully three-dimensional, realistic model consisting of the MHV plus aortic root, mimicking the physical model embedded in a mock loop (Scotten et al., 2002; Grigioni et al., 2003; Redaelli et al., 2004), was implemented. A previously developed implicit FSI method (Nobili et al., 2007) was parallelized (16 nodes) and DNS was performed. The ALE method was used for modelling the leaets and uid motion. Indeed the ALE approach, which combines the use of the classical Lagrangian and Eulerian reference frames, is used largely in the analysis of FSI systems (Donea et al., 2004), with the inclusion of MHVs (Cheng et al., 2004). The results of the dynamics of the valve were validated with an experimental counterpart, measuring valve leaet kinematics in vitro with the ultrafast cinematographic technique (Barbaro et al., 1997).
2. Materials and methods
The kinematics of a St. Jude bileaet MHV (27 mm tissue annulus diameter, maximum opening angle 851) was investigated using both experimental and computational approaches. In particular, a pulsatile, open loop mock circulatory system (MCS) was used, properly modied in

ARTICLE IN PRESS
M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 order to allow visualization of the valvular function. In parallel, numerical simulations were carried out in a computational domain designed to closely mimic the in vitro experimental setup. In both cases, the valve dynamics was simulated in physiological ow condition and with realistic geometry (Valsalva sinus, valve hinges). 2541

2.1. Experimental method

The MCS-VSI (Vivitro Systems Inc., Canada, Fig. 1) is equipped with a piston-in-cylinder pump head pulse duplicator, and a physical model of the left heart system made of: (i) a hydraulic chamber containing a transparent and compliant polyurethane sac mimicking the left ventricle; (ii) an open reservoir that mimics the atrial chamber; (iii) a simplied glass model of the aortic root, with three hemispheres placed at 1201 symmetry mimicking the sinuses of Valsalva. A detailed description of the MCS can be found in Grigioni et al. (2003) and in Redaelli et al. (2004). A mechanical no-leakage valve was inserted in the mitral site of the physical model. This was done in order to avoid interferences in the dynamics of the aortic MHV under investigation, due to the dynamics of the mitral valve (not present in the in silico model). A monitoring system (accurately described in Redaelli et al., 2004, Grigioni et al. 2003) allowed acquiring and analyzing ventricular, atrial, and aortic pressure, together with aortic and mitral ow rate. The MCS was driven by imposition of a pump drive waveform resembling an instantaneous ventricular volume. A driving waveform corresponding to physiological ow conditions (a cardiac output of 4.5 l/min) was set (Fig. 2). A mean arterial pressure of 100 mmHg and a heart beat of 72 beats per minute (bpm) was prescribed. A blood analogue, 35% waterglycerol solution (3.7 103 Pa s dynamic viscosity), was used as test uid. The study of leaet kinematics was performed using ultrafast cinematographic technique, which allows recording the movement of characteristic points of the leaets during the working cycle of the MHV. To reach this aim, the mock circulatory loop was modied to allow vision of the valvular function in the aortic site by means of a thin glass window (to guarantee against optical distortions) at the top of the aortic site, through which a high-speed video camera was oriented (Kodak Ektapro, which records 239 192 pixels images at a frequency fc 1000 frames/s). Time resolution was gained by splitting the camera screen into 12 parallel slices, thus grabbing the image of the central piece of the prosthetic valve at a rate of 12,000 frames/s. This allowed recording the position of the

Fig. 2. Driving waveform corresponding to physiological ow condition applied to experimental and numerical simulations.

innermost points at the internal left and right leaet edge, thus measuring the width of the gap between the leaets (B Datum): the knowledge of this sole measured component of the displacement is sufcient to fully describe leaet dynamics for St. Jude HP valve model, due to the design features of the coupling between the housing ring and the leaet, which allows only a pure rotational motion to the leaet (rigid motion). In this way, it was possible to follow in time leaet motion in detail. For each experimental section, 10 cycles were acquired and averaged (Ektapro recording synchronized with the cardiac cycle). Exhaustive details about the leaet motion analysis can be found in Barbaro et al. (1997) and Redaelli et al. (2004).

2.2. Numerical simulation

2.2.1. Numerical model A detailed three-dimensional model was shaped on the geometry of the experimental setup (Fig. 3). The aortic and ventricular conduits were constructed from the two-dimensional MCS technical drawings. Only the region located in between the sections hosting the MCS pressure sensors was taken into account (domain of interest) and, as in the MCS, the Valsalva sinuses were modelled as three semispheres (1201 symmetry) placed downstream of the valve site (Fig. 3). The geometry of the valve, including the valve housing and the hinges geometry, was constructed from data provided by St. Jude Medical. Leaets were scaled by a factor 0.98, thus allowing us to put three rows of cells in between the closed leaets and the valve housing, and to attain an high-quality mesh during the whole cardiac cycle. To minimize the inuence from outlet boundary conditions, a straight ow extension (10 cm length) was added 14.5 cm downstream of the aortic duct. The valve region was discretized with tetrahedral elements according to the moving deforming mesh module requirements (described in the following section), and the mesh was locally rened to accommodate for the leaet movement. In the remaining domain (aortic and ventricular ducts) a hexahedral mesh was used in order to limit the numerical diffusion. The total number of cells used in the simulation was equal to 2.1 millions (1.2 milion tetrahedral and 900,000 hexahedral). The dimension of the grid was set according to the assumption that large and intermediate eddies are the energy containing eddies (Kolmogorov, 1941a, b), and are the structures responsible for leaet motion, Hence, we set a grid density that allows for solving the Taylor microscale (calculated at peak systole, considering the diameter of the

Fig. 1. Schematic of the pulsatile, open loop, mock circulatory system (MCS)-VSI (Vivitro Systems Inc., Canada). The MCS consists of a pump system (PS) controlled by PC for waveform generation and data acquisition and a physical model of the left heart system. The left ventricle chamber (LVC) consists of a left ventricle model made of polyurethane, a glass model for the aortic root, and an entrapping air system for the aortic compliance. The atrium consists of an open ceiling reservoir (OCR). The left ventricle and the atrium are connected by a circuit with a variable peripheral resistance (VPR). The St. Jude HP valve is located in the aortic valve site (AVS); the aortic ow measurements were performed by an electromagnetic ow meter (FM).

ARTICLE IN PRESS
2542 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550

Fig. 3. Conguration of the computational domain: (a) zoom into the St Jude valve site and (b) top view of the valve in closed position that clarify the orientation of the leaet with respect to the Valsalva sinuses (black arrow pointing to the leaet tip).

straight conduit at the sinutubular region as the integral length scale), which represents the length scale falling in between the large scale eddies and the small scale eddies (further details in the Discussion section). The grid density was increased around the leaets and in the whole aortic root, where the three jet-like congurations of the velocity eld and the vortex shedding phenomena characterize the ow eld. The models and the mesh generation were performed by using software GAMBIT (Ansys Inc., USA). The simulations were performed adopting a control-volume-based technique (nite-volume method) to solve the NavierStokes equations (FLUENT general-purpose uid dynamic code, Ansys Inc., USA). To match the rheology of the test uid used in the experimental campaign, blood was modelled as an isotropic, homogenous, incompressible and Newtonian uid (with a 0.0037 Pa s viscosity). The aortic walls were assumed rigid, and no slip (i.e., zero velocity) conditions were set at the walls. At the inlet, we assigned a ow rate that corresponds to the one prescribed in the experimental section, with the inow boundary condition set in terms of velocity proles: a at inow velocity prole for the axial velocity and zero transverse velocity components were assigned at the ventricular duct inlet. To solve the ow eld we applied a DNS. In this approach the ow eld is solved directly from the NavierStokes equations and no averaging or turbulence modelling is applied. The cardiac cycle (0.8 s) was discretized with 4000 time steps, corresponding to a physical time step of 0.2 ms. Two cycles were simulated in order to remove the dependence of the opening phase from the initial condition and guarantee cycle invariance. 2.2.2. Fluid structure interaction An accurate solution of the Navier Stokes equations for deforming meshes is provided by the use of the ALE formulation, which makes it possible to include grid velocities in the momentum and continuity equation of the uid domain (Le Tallec and Mouro, 2001). The ALE description conjugates Lagrangian and Eulerian features. The computational grid is neither moved with the boundary (Lagrangian) or held xed (Eulerian). Rather, it is moved in some arbitrarily specied way to give a continuous reconguration capability. Because of this freedom in moving the computational mesh offered by the ALE description, greater distortions of the continuum can be handled better than would be allowed by a purely Lagrangian method, with more resolution than that afforded by a purely Eulerian approach. The partitioned approach was used to simulate the interplay between leaets and blood. This strategy preserves the uid and the structural solvers as separate. Both parts are alternately integrated in time and the interaction is taken into account by the boundary conditions of both the solvers. As a

direct consequence there exists an intrinsic time lag between the integration of the uid and the structure, which can be avoided by repeating the interaction until both the solution consistently produce the same result. The general scheme of the coupling procedure is shown in Fig. 4. As mentioned above, the uid domain was solved using the nitevolume method computational code Fluent (Ansys Inc., USA), which provides a number of features well suited to handle the specic problem of rotating boundaries. We used a spring-based moving, deforming mesh module, which allows a robust mesh deformation handling by assuming that the mesh element edges behave like an idealized network of interconnected springs. In order to maximize the inuence of the boundary node displacements on the motion of the interior nodes, no damping was applied to the springs (spring constant factor 0). To preserve the quality of the mesh during the valve motion, the maximum admissible skewness of the computational cells was set equal to 0.7. The Fluent remeshing algorithm was adopted to properly treat degenerated cells, which agglomerates cells that violate the skewness criterion, and locally remeshes the agglomerated cells. If the new cells satisfy the skewness criterion, the mesh is locally updated with the new cells (with the solution interpolated from the old cells); otherwise, the new cells are discarded (FLUENT Users Manual, 2007). In this work, the moving deforming mesh module was used in conjunction with two user-dened subroutines, named MDM and center of gravity motion (cg_motion), respectively; at the beginning of each step the rst one calculates and updates the kinematics of the leaets on the basis of the moment applied to the leaet, which is calculated by the second subroutine at the end of the time step, once the time step convergence has been achieved. An iterative call to the uid solver is performed by an external subroutine in order to update the solution of the uid dynamic eld and achieve the convergence of the FSI cycle, until the difference between the external momentum divided by the inertia of the uid (calculated by MDM) and the angular acceleration (imposed by cg_motion) is not below a threshold value (e 500 s2). More in detail, since the valve leaet is rigid body in rotation on a xed pivot, the angular position is the only degree of freedom and leaet dynamics can be calculated using: Mp Mt IW the angular where I is the angular inertia, equal to 8.75 109 Kg m2, W acceleration of the leaet, Mp the torque applied on the leaet external surface by the pressure eld, and Mt is the moment generated by shear stresses. The acceleration value for the subsequent iteration within the generic time step i is updated through an under-relaxation scheme (Le Tallec and

ARTICLE IN PRESS
M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 2543

Fig. 4. Implicit coupling procedure scheme of the interaction between the valve leaet and the blood ow: W is the angular position of the leaet position, i the time step number, k the number of FSI iteration within the same time step, o the under-relaxation coefcient, and e the FSI convergence threshold. Two loops are annealed: the loop that checks for the convergence of the FSI procedure (inner loop), and the temporal advancement loop that makes the estimation ). simulation proceed to the next time step starting from the acceleration of the leaet calculated at the end of the previous one (W

Mouro, 2001) as k w k1 W W i i
1 k1 Mk pi M tii

k W i

where k is the iteration index and w (set equal to 0.01) is the underrelaxation factor, which plays the role of damping changes in the acceleration produced during each iteration. Starting from the acceleration obtained in Eq. (2), the velocity and the displacement of the leaet were calculated using the Newmark method. More details about the coupling procedure can be found in Nobili et al. (2007).

3. Results Fig. 5a shows the angular valve displacement obtained from numerical and experimental simulations. For the sake of clarity, among the measured cycles (10 cycles), four of them are reported as representative of the whole data set (Fig. 5a). At the early acceleration phase, the experimental measurements put in evidence a highly repeatable kinematics of the leaets. Accordingly, during the opening phase, the average of the displacement of the leaets over all the measured cycles is representative of the phenomenon under investigation. The numerical results showed a maximum delay, with respect to the average experimental curve, of about 4.1 ms (within the 72SD) after about 8 ms. Afterward the experimental variability in the leaets

position increases and becomes particularly pronounced during the valve closure. This different behavior exhibited by the MHV at closure in vitro, under the same experimental conditions, is widely documented in literature and could be ascribed to the unsteadiness of mock loops, or to cyclic variations existing in the pulsatile ow (Kleine et al., 1998). In order to make a consistent comparison between numerical and experimental results, the mean closing time was used. The experimental simulations displayed a mean closing time equal to 0.3324 s (Texp), a value that is very close to the numerical result of 0.34 s (Tnum), which again falls in a range of two times the standard deviation (std 0.0058 s). The comparison between experimental and numerical transvalvular pressure drop waveforms is depicted in Fig. 6. In the numerical simulation, aortic and ventricular pressures were calculated by integrating pressure values at sections corresponding to the ones where pressure was measured in the experimental session. Experimental and numerical transvalvular pressure drops agree satisfactorily. Differences between them can be ascribed to the role played during the experiments by the systemic and the aortic root compliance, which have not been taken into account in the in silico model. Hence, the numerical model represents a system more rigid than the

ARTICLE IN PRESS
2544 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550

Fig. 5. Comparison between experimental and numerical results: (a) angular displacement and (b) zoom to the opening phase.

Fig. 6. Comparison of numerical and experimental pressure drops through the valve during the systolic phase. Pressure was evaluated at a distance of 2.5 diameters upstream and downstream of the valve site.

experimental one. This speculation is sustained by the fact that the numerical simulation put in evidence maximum absolute values for the numerical transvalvular pressure drop, higher than that of the experimental setup (about 6.6 mmHg difference). The temporal evolution of the axial velocity through the valve is shown in Fig. 7. Four time instants representative of the cardiac cycle are considered, referring to the central longitudinal plane. In accordance with consolidated results on St. Jude MHVs (see for example Yoganathan et al., 2004) a triple jet pattern characterizes the forward ow, the major part of which emerges from the two side orices: the growing magnitude of the three jets outgoing from the valve during the acceleration phase is clearly visible as well as the recirculation areas in the sinus region and the evolution of the vortices shed by the leaets. The maximum value of 1.38 m/s is observed at peak systole (Fig. 7 panel 2) in the jets emerging from the side orices of the prosthetic device.

This can be better appreciated in Fig. 8, depicting (magnied view of the ow eld, with respect to Fig. 7) the velocity vector eld superimposed on the velocity color contour at peak systole. During the deceleration phase, the ow is more evenly distributed downstream of the valve plane (Fig. 7 panel 3), and it emerges from central orice with a speed higher than during the acceleration phase. This loss of organization in the ow eld during the deceleration phase is due to inertial effects, in a manner that is typical for intermittent turbulent ows in the transitional range (Yin et al., 2004). The closing phase (Fig. 7 panel 4) takes place when the ow is completely inverted. Fig. 9 shows the vorticity distribution in the same transversal plane used for the axial velocities and for the same instants. As the valve opens and the ow rate grows, the vorticity magnitude increases in the valve region. The maximum values of vorticity are reached all around the leaet surface, near the valve ring, and in the transitional areas between the three jets. The presence of a recirculating region, pointed out in the vorticity map, is clearly evident within the sinus of Valsalva and in the opposite location, in accordance with the inversion of the velocity values in this region, depending on design features of the MHVs and also on the shape of the aortic root geometry. In order to elucidate the intricate structure of the pulsatile ow through the valve, the computed results are analyzed from a Lagrangian viewpoint. Particle traces can offer a four-dimensional (both in space and time) visualization of ow patterns and have proven to be useful tools to interrogate complex ow elds in vessels (Markl et al., 2004; Steinman, 2000; Morbiducci, et al., 2007) and, in general, to reveal global organization of blood ow (Buonocore and Bogren, 1999). Hence, a framework of the evolving ow eld is obtained by emitting massless particles 2 mm upstream of the valve site. Rendered trajectories, color coded with the residence time (starting from the instant of injection, at peak systole) are shown in Fig. 10, where the particle traces selected on the basis of their high helical content (according to Grigioni et al.,

ARTICLE IN PRESS
M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 2545

Fig. 7. Sequential presentation of the axial velocity contour plot at four different instants of the cardiac cycle: opening phase (1); peak systole (2); decelaration phase (3); and closing phase (4).

2005b and Morbiducci et al., 2007, who recently used helicity-related quantities to depict important topological features of blood ow) are depicted: the panel A allows us to appreciate the highly three-dimensional structure of the computed ow eld. In particular, the Lagrangian visualization put in evidence (Fig. 10, panel B) three-dimensional vortices with higher helical content in sinus II and sinus III (due to the orientation of the leaet), and helical structures in ow channels developing downstream of each sinus (arrowheads). 4. Discussion In recent years, due to growing computational power, and the need to investigate with greater accuracy the hemodynamic performance of prosthetic heart valves, the use of FSI modelling has been introduced in the numerical simulations of these devices. In this study we included the fully nonlinear FSI effects between the leaet motion and blood ow, use of realistic geometry and experimental validation of leaet dynamics. Many improvements were done with respect to our previous work (Redaelli et al., 2004). Indeed, the investigation

Fig. 8. Velocity vector plot of the ow past the valve at peak systole (midplane cross-section).

ARTICLE IN PRESS
2546 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550

Fig. 9. Contour plot of the vorticity magnitude at ve different instants of the cardiac cycle: opening phase (1); peak systole (2); decelaration phase (3); and closing phase (4).

Fig. 10. Three-dimensional visualization of the traces of the particles emitted on a surface 2 mm upstream of the valve site (panel A). The rendered trajectories are color coded with the residence time (starting from the instant of injection, at peak systole), and selected in terms of their helical content (Morbiducci et al., 2007). The Lagrangian visualization put in evidence (panel B) three-dimensional vortices with higher helical content in sinus II and sinus III (due to the orientation of the leaet), and helical structures in ow channels developing downstream of each sinus (arrowheads).

was extended from the opening phase to the whole systole and the ow rate was increased to physiological values. Concerning the numerical model, the simplications at the geometry of the hinges and at the conguration of the Valsalva sinus were removed, thus leading to a more

detailed evaluation of the behavior of the valve. Moreover, the coupling between leaets and blood was changed from explicit in time to implicit in order to remove the time delay in leaet response due to the intrinsic numerical limits of the explicit scheme. As recently stated (Nobili et al., 2007),

ARTICLE IN PRESS
M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 2547

implicit coupling suppresses physical inaccuracy at small space/time scale or, equivalently at high derivatives. To validate the numerical prediction of the dynamics of the leaets, we compared the computed leaet displacement with the experimental data obtained from a MCS. The remarkable agreement between leaet displacement, shown in Fig. 2, demonstrates the capability of implicit coupling proposed herein to capture the interaction between the leaet movement and blood ow. The valve opening motion was notably consistent from cycle to cycle, perhaps due to the stable nature of accelerating ow. On the contrary, leaet kinematics measurements show that the position of the leaets exhibits a more pronounced variability from cycle to cycle during the valve closure phase: our experimental results are similar to the highly time resolved PIV measurements by Kaminsky et al. (2007a) and Dasi et al. (2007). Compared to this study last we moved in a different direction. We checked for a spatial grid sufciently rened to reproduce the in vitro observed leaet dynamics, modelling a more realistic geometry for the aortic root and for the valve, and choosing a real FSI approach to the problem. Thereby, the over 2 millions of elements used to discretize the computational domain represent a tradeoff between accuracy and computational costs. However, Fig. 510 clearly testify that the proposed numerical scheme is able to capture the relevant structures in the ow eld downstream of the MHV, in accordance with the state-of-art in silico and in vitro observations (Browne et al., 2000; Grigioni et al., 2000; Yoganathan et al., 2004, 2005; Dumont et al., 2007; Kaminsky et al., 2007b). Temporal and local information on the axial velocity and vorticity magnitude of the blood owing through the MHV was obtained in the current study. The presence of three sinuses of Valsalva in the aortic root makes the region immediately downstream of the aortic valve nonsymmetric, with remarkable consequences in the ow domain. The rendered trajectories elucidate the rich three-dimensional structure of the ow eld: ow visualization based on trace selection was applied to permit extraction of salient ow features data simplication, and compression. The current unsteady numerical calculation has high computational costs. For the completion of calculation for one cardiac cycle, 790 h on a single processor (Xeon 3.2 GHz, 8Gb RAM) were taken. The use of parallel computing has considerably reduced the computational costs, showing a linear scaling up to 16 processors as previously observed by other authors (Yue et al., 2004; Wang et al., 2007) despite the use of external subroutine to manage the FSI process. Although numerical computational techniques are powerful tools in the design and virtual assessment of MHVs, they still have limitations. The presence of a gap several orders of magnitude smaller than the aorta diameter (1.118 mm vs 27 mm ) represent a remarkable problem for the discretization of the computational

domain (high cell size variation and large cell number needed) that actually could not be handled by the CAE software. Moreover, FSI requires physical separation of the uid domain from the structural domain, dictating the inclusion of a gap between the closed valve leaets and their housing. Accordingly, the clearance gaps were achieved by slightly reducing the size of the leaets to 98% of their actual size, yielding a central gap of 98.633 mm (original size 1.118 mm). This leads to an overestimation of the leakage ow through the gaps and the consequent underestimation of the related shear stress levels, as pointed out by Dumont et al. (2007). Hence, the leakage ow was not analyzed in the present work. However, scaling leaet dimensions could directly affect their motion due to the reduced value of moment of inertia (the angular position of the leaet is a function of its moment of inertia, see Eq. (1)). The consequence for this could be that the dynamics of scaled leaets could be different from the real ones. The values of the moment of inertia used in our study for the scaled leaet is equal to 8.75 109 kg m2, and it was calculated using a reference value for the density of 2200 kg m3, according to the data provided by St. Jude Medical for pyrolitic carbon. The moment of inertia of the leaets of real dimension, for the same density value given by the manufacturer, is equal to 9.94 109 kg m2. The application of a scaling factor of 0.98 leads also to a 4% leaet total area reduction. Hence, the 98% of leaet scaling, even if only 98%, is likely partially responsible for the differences in leaet dynamics between numerical and experimental results, in particular during the opening phase (Fig. 5 left panel): after the rst 10 ms (where the inertial effect of the uid column impinging on the valve is predominant), the numerical leaet dynamics is more rapid than the real leaet kinematics, in consequence of the lower moment of inertia of the modelled leaets. The role played by the lowered moment of inertia for leaets dynamics in silico at closure is less clear, due to the variability in the closing behavior exhibited by the valve in the experimental setup (Fig. 5 right panel). Concerning the full nonlinear FSI, uncertainties due to lack of specic knowledge of two important parameters affecting the motion of the leaets were undertaken. In particular, we neglected modelling the compliance of the aortic root (which contributes to the difference observed in transvalvular pressures), which is generally modelled in vitro by an entrapping air system in the physical model representing our experimental counterpart, and the friction forces due to the presence of the hinge mechanism (idealized in our model), which is valve specic and not easy to assess. Notwithstanding such parameters not being taken into account in the simulation, the modelled leaets dynamics resembles the cycle-to-cycle variation measured in vitro, as shown in Fig. 4. As previously mentioned, during the construction of the computational grid for DNS implementation, we aimed at a tradeoff between accuracy and computational costs.

ARTICLE IN PRESS
2548 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550

Actually the resolution of the grid is another limitation of the present investigation. The range of scales that need to be accurately represented in a computation is dictated by the physics. According to common literature on MHV uid dynamics, we considered the diameter of the straight conduit at the sinutubular region (D 27 mm) as the length scale of larger eddies in the investigated ow eld, characterized by a maximum Reynolds number value equal to 6000, and a mean Reynolds number value equal to 3775. Hence, the Kolmogorov scale Z D Re3/4 of the investigated ow eld results in the order of 0.04 and 0.06 mm for the maximum and the mean Reynolds, respectively. This represents an important indication for the dimension of the grid, although it has been demonstrated that the achievement of dissipative length scale in terms of grid size is not needed to have an accurate solution (Moin and Mahesh, 1998; Moser & Moin, 1987; Spalart, 1988; Rogers et al., 1987). Indeed, the numerical grid determines the scales that are represented in the simulation: while in the gap region and close to the surfaces of the leaet we set a computational cell mean dimension equal to 0.03 mm, the spatial discretization inside the straight conduit downstream of the sinutubular region is 0.49 mm, i.e., one order magnitude greater than the Kolmogorov scale. This choice is dictated, since DNS, even though is the most accurate way of simulating turbulent ow, unfortunately, is also the most expensive way. Hence, the prediction of such ows at a reduced computational cost with a suitably developed turbulence model is a widely used practice in studying ows through prosthetic valves. There is a great variety of turbulence models available through commercial codes for computational uid dynamics, especially the ones employing the Reynolds averaged NavierStokes (RANS) approach. However, these twoequation models suffer from severe limitations. As also pointed out by Varghese et al., (2008), most of these turbulence models have been developed using knowledge of simple classes of well-behaved two-dimensional ows. In virtue of the computational resources needed for a realistic FSI simulation of the turbulent ow through a prosthetic valve, we beg the question whether it is appropriate to consider a grid size greater than the Kolmogorov length scale to capture the effects of features of a through-mechanical valve ow eld involved in leaet dynamics. Features that can be thought as the ones characterizing the integral scale of the ow. Very recently Dasi et al. (2007) used the DNS approach, which is the same approach adopted in the present investigation, to depict the transient ow past a MHV and the wake dynamics. Dasi and colleagues were able to use an highly resolved spatial grid (about 9.7 millions of nodes), which solves the Kolmogorov scale. However, the price to be paid by Dasi et al. (2007) was neglecting the FSI and oversimplifying the geometry. The good agreement of the results obtained in our study with experiments in terms of leaet dynamics supports the

hypothesis that the dynamics of the leaets is mainly dependent on the integral scale of the ow, because even though the Kolmogorov length scale is not resolved (i.e., turbulence is not sustained at the dissipative scale), the ow eld is resolved down to the inertial subrange, where motions are determined by inertial effects. Accordingly, the relevant length scales would go from the integral one down to the Taylor microscale l D(10/Re)1/2, which falls in between the large scale eddies and the small scale eddies and is equal to 1.1 mm for the investigated ow eld, at peak Reynolds (i.e., the double of the cell size downstream of the sinutubular region). We believe that the FSI approach, coupled with the decision to go the DNS route, even if not fully resolving the Kolmogorov scale at this stage of the investigation, is promising. In the future, we plan to combine the quantitative depiction of ow elds and stresses within the MHV with blood damage accumulation models (Grigioni et al., 2005c; Nobili et al., 2008) in order to evaluate the hemolytic and the thrombogenic potency of the MHV. Moreover we are currently working on a more comprehensive analysis of the helical ow structure downstream of the valve by using the helical ow index proposed by Morbiducci et al. (2007). These calculations will require a ner computational mesh compared to the one used in the present study (increasing the grid density, thus moving toward the Kolmogorov scale) in order to enhance resolution in the region downstream of the leaets, where a chaotic ow emerges in the second part of the systole, as very recently assessed both in vitro and in silico by Dasi et al. (2007). In conclusion, the favorable agreement obtained between in vitro and numerical results of the leaet displacements suggests the application of FSI models as a major tool to investigate and/or solve problems related to the implantation of MHVs, and to optimize their design. Conict of interest Disclosures and freedom of investigation The research on FSI applied to prosthetic heart valves was not supported nancially by the St. Jude company. The devices were not donated by the company; they were acquired at the regular market cost in Italy. None of the authors has a nancial agreement with the company St Jude. None of the authors has conicts of interests in the study. All authors had full control of the design of the study, methods used, outcome parameters, analysis of the data, and production of the written report. References
Alemu, Y., Bluestein, D., 2007. Flow-induced platelet activation and damage accumulation in a mechanical heart valve: numerical studies. Articial Organs 31 (9), 677688.

ARTICLE IN PRESS
M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 Barbaro, V., Grigioni, M., Daniele, C., Boccanera, G., 1997. Reconstruction of closing phase kinematics by motion analysis for a prosthetic bileaet valve. In: Power, H., Brebbia, C.A., Kenny, JM. (Eds.), Simulations in Biomedicine IV. Computational Mechanics Publications. WIT Press, Southampton, Boston, pp. 349358. Bluestein, D., Rambod, E., Gharib, M., 2000. Vortex shedding as a mechanism for free emboli formation in mechanical heart valves. Journal of Biomechanical Engineering 122 (2), 125134. Browne, P., Ramuzat, A., Saxena, R., Yoganathan, A.P., 2000. Experimental investigation of the steady ow downstream of the St. Jude bileaet heart valve: a comparison between laser Doppler velocimetry and particle image velocimetry techniques. Annals of Biomedical Engineering 28 (1), 3947. Buonocore, M.H., Bogren, H.G., 1999. Analysis of ow patterns using MRI. International Journal of Cardiac Imaging 15, 99103. Cheng, R., Lai, Y.G., Chandran, K.B., 2004. Three-dimensional uidstructure interaction simulation of bileaet mechanical heart valve ow dynamics. Journal of Heart Valve Disease 12, 772780. Dasi, L.P., Ge, L., Simon, H.A., Sotiropoulos, F., Yoganathan, A.P., 2007. Vorticity dynamics of a bileaet mechanical heart valve in an axisymmetric aorta. Physics of Fluids 19 (067105), 17. DeWall, R.A., Qasim, N., Carr, L., 2000. Evolution of mechanical heart valves. Annals of Thoracic Surgery 69 (5), 16121621. Donea, J., Huerta, A., Ponthot, J.-Ph., Rodriguez-Ferran, A., 2004. Arbitrary LagrangianEulerian methods. In: Erwin, Stein, Ren0 e de Borst, Thomas, J.R., Hughes (Eds.), Fundamentals. Encyclopedia of Computational Mechanics, vol. 1. Wiley, New York. Dumont, K., Vierendeels, J.A., Segers, P., Van Nooten, G.J., Verdonck, P.R., 2005. Predicting ATS open pivot heart valve performance with computational uid dynamics. Journal of Heart Valve Disease 14 (3), 393399. Dumont, K., Vierendeels, J., Kaminsky, R., van Nooten, G., Verdonck, P., Bluestein, D., 2007. Comparison of the haemodynamic and thrombogenic performance of two bileaet mechanical heart valves using a CFD/ FSI model. Journal of Biomechanical Engineering 129 (4), 558565. Fluent Inc., 2007. FLUENT Users Manual, 2007. FLUENT Software Package, ver. 6.3.26, Fluent Inc., Lebanon, NH. Ge, L., Jones, S.C., Sotiropoulos, F., Healy, T.M., Yoganathan, A.P., 2003. Numerical simulation of ow in mechanical heart valves: grid resolution and the assumption of ow symmetry. Journal of Biomechanical Engineering 125 (5), 709718. Gott, V.L., Alejo, D.E., Cameron, D.E., 2003. Mechanical heart valves: 50 years of evolution. Annals of Thoracic Surgery 76 (6), S2230S2239. Grigioni, M., Daniele, C., D0 Avenio, G., Barbaro, V., 2000. On the monodimensional approach to the estimation of the highest Reynolds shear stress in a turbulent ow. Journal of Biomechanics 33 (6), 701708. Grigioni, M., Daniele, C., Romanelli, C., Morbiducci, U., D0 Avenio, G., Del Gaudio, C., Barbaro, V., 2003. Pathological patient in protocol denition for bench testing of mechanical cardiac support systems. International Journal of Articial Organs 26 (1), 6472. Grigioni, M., Daniele, C., D0 Avenio, G., Morbiducci, U., Del Gaudio, C., Abbate, M., Di Meo, D., 2004. Innovative technologies for the assessment of cardiovascular medical devices: state of the art techniques for articial heart valves testing. Expert Review of Medical Devices 1 (1), 89101. Grigioni, M., Daniele, C., Del Gaudio, C., Morbiducci, U., Balducci, A., DAvenio, G., Barbaro, V., 2005a. Three-dimensional numerical simulation of ow through an aortic bileaet valve in a realistic model of aortic root. Asaio Journal 51 (3), 176183. Grigioni, M., Daniele, C., Morbiducci, U., Del Gaudio, C., DAvenio, G., Balducci, A., Barbaro, V., 2005b. A mathematical description of blood spiral ow in vessels: application to a numerical study of ow in arterial bending. Journal of Biomechanics 38 (7), 13751386. Grigioni, M., Morbiducci, U., DAvenio, G., Di Benedetto, G., Del Gaudio, C., 2005c. Proposal for a new formulation of the power law mathematical model for blood trauma prediction. Biomechanics and Modeling in Mechanobiology 4 (4), 249260. 2549 Guivier, C., Deplano, V., Pibarot, P., 2006. New insights into the assessment of the prosthetic valve performance in the presence of subaortic stenosis through a uidstructure interaction model. Journal of Biomechanics 40 (10), 22832290. Kaminsky, R., Morbiducci, U., Rossi, M., Scalise, L., Verdonck, P., Grigioni, M., 2007a. Time resolved PIV technique allows a high temporal resolution measurement of mechanical prosthetic aortic valves uid dynamics. International Journal of Articial Organs 30 (2), 153162. Kaminsky, R., Dumont, K., Weber, H., Schroll, M., Verdonck, P., 2007b. PIV validation of bloodheart valve leaet interaction modelling. International Journal of Articial Organs 30 (7), 640648. Kelly, S., 2002. Computational uid dynamics insights in the design of mechanical heart valves. Articial Organs 26 (7), 608613. Kleine, P., Perthel, M., Nygaard, H., Hansen, S.B., Paulsen, P.K., Riis, C., Laas, J., 1998. Medtronic Hall versus St. Jude Medical mechanical aortic valve: downstream turbulences with respect to rotation in pigs. Journal of Heart Valve Disease 7, 548555. Kolmogorov, A.N., 1941a. Local structure of turbulence in an incompressible uid for very large Reynolds numbers. Doklady Akademii Nauk SSSR 30, 299303. Kolmogorov, A.N., 1941b. Energy dissipation in locally isotropic turbulence. Doklady Akademii Nauk SSSR 32, 1921. Lai, Y.G., Chandran, K.B., Lemmon, J.A., 2002. Numerical simulation of mechanical heart valve closure uid dynamics. Journal of Biomechanics 35 (7), 881892. Le Tallec, P., Mouro, J., 2001. Fluid structure interaction with large structural displacements. Computer Methods in Applied Mechanics and Engineering 190, 30393067. Leo, H.L., Simon, H.A., Dasi, L.P., Yoganathan, A.P., 2006. Effect of hinge gap width on the microow structures in 27 mm bileaet mechanical heart valves. Journal of Heart Valve Disease 15 (6), 800808. Leo, H.L., He, Z., Ellis, J.T., Yoganathan, A.P., 2002. Microow elds in the hinge region of the CarboMedics bileaet mechanical heart valve design. Journal of Thoracic and Cardiovascular Surgery 124 (3), 561574. Markl, M., Draney, M.T., Hope, M.D., Levin, J.M., Chan, F.P., Alley, M.T., Pelc, N.J., Herfkens, R.J., 2004. Time-resolved 3-dimensional velocity mapping in the thoracic aorta. Visualization of 3-directional blood ow patterns in healthy volunteers and patients. Journal of Computer Assisted Tomography 28, 459468. Moin, P., Mahesh, K., 1998. Direct numerical simulation: a tool in turbulence research. Annual Review of Fluid Mechanics 30, 539578. Morbiducci, U., Ponzini, R., Grigioni, M., Redaelli, A., 2007. Helical ow as uid dynamic signature for atherogenesis in aortocoronary bypass. A numeric study. Journal of Biomechanics 40 (3), 519534. Moser, R.D., Moin, P., 1987. The effects of curvature in wall-bounded turbulent ows. Journal of Fluid Mechanics 175, 479510. Nobili, M., Passoni, G., Redaelli, A., 2007. Two uidstructure approaches for 3D simulation of St. Jude Medical bileaet valve opening. Journal of Applied Biomaterials and Biomechanics 5 (1), 4959. Nobili, M., Sheriff, J.F., Morbiducci, U., Redaelli, A., Jesty, J., Bluestein, D., 2008. Platelet activation due to hemodynamic shear stresses: damage accumulation model and comparison to in vitro measurements. Asaio Journal 54 (1), 6472. Pelliccioni, O., Cerrolaza, M., Herrera, M., 2007. Lattice Boltzmann dynamic simulation of a mechanical heart valve device. Mathematics and Computers in Simulation 75 (12), 114. Penrose, J.M.T., Staples, C.J., 2002. Implicit uidstructure coupling for simulation of cardiovascular problems. International Journal for Numerical Methods in Fluids 40 (3-4), 467478. Redaelli, A., Bothorel, H., Votta, E., Soncini, M., Morbiducci, U., Del Gaudio, C., Balducci, A., Grigioni, M., 2004. 3-D simulation of the SJM bileaet valve opening process: uidstructure interaction study and experimental validation. Journal of Heart Valve Disease 13, 804813.

ARTICLE IN PRESS
2550 M. Nobili et al. / Journal of Biomechanics 41 (2008) 25392550 of ow instabilities in anatomical total cavopulmonary connections. Annals of Biomedical Engineering 35 (11), 18401856. Yin, W., Alemu, Y., Affeld, K., Jesty, J., Bluestein, D., 2004. Flowinduced platelet activation in bileaet and monoleaet mechanical heart valves. Annals of Biomedical Engineering 32 (8), 10581066. Yoganathan, A.P., He, Z., Casey Jones, S., 2004. Fluid mechanics of heart valves. Annual Review of Biomedical Engineering 6, 331362. Yoganathan, A.P., Chandran, K.B., Sotiropoulos, F., 2005. Flow in prosthetic heart valves: state-of-the-art and future directions. Annals of Biomedical Engineering 33 (12), 16891694. Yokoyama, Y., Medart, D., Hormes, M., et al., 2006. CFD simulation of a novel bileaet mechanical heart valve prosthesis: an estimation of the Venturi passage formed by the leaets. International Journal of Articial Organs 29 (12), 11321139. Yue, X., Hwang, F.N., Shandas, R., Cai, X.C., 2004. Simulation of branching blood ows on parallel computers. Biomedical Sciences Instrumentation 40, 325330. Rogers, M.M., Moin, P., 1987. The structure of the vorticity eld in homogeneous turbulent ows. Journal of Fluid Mechanics 176, 3366. Scotten, L.N., Walker, D.K., Dutton, J.W., 2002. Modied Gorlin equation for the diagnosis of mixed aortic valve pathology. Journal of Heart Valve Disease 11, 360368. Spalart, P.R., 1988. Direct numerical simulation of a turbulent boundary layer up to Rey 1410. Journal of Fluid Mechanics 187, 6198. Steinman, D.A., 2000. Simulated pathline visualization of computed periodic blood ow patterns. Journal of Biomechanics 33 (5), 623628. Varghese, S.S., Frankel, S.H., Fischer, P.F., 2008. Modeling transition to turbulence in eccentric stenotic ows. Journal of Biomechanical Engineering 130 (1), 014503. Vierendeels, J., Dumont, K., Dick, E., Verdonck, P., 2005. Analysis and stabilization of uidstructure interaction algorithm for rigid-body motion. AIAA Journal 43 (12), 25492557. licourt, D., Horner, M., Parihar, A., Wang, C., Pekkan, K., de Ze Kulkarni, A., Yoganathan, A.P., 2007. Progress in the CFD modeling