ORIGINAL RESEARCH published: 22 March 2019 doi: 10.3389/fphys.2019.00237 In silico Optimization of Left Atrial Appendage Occluder Implantation Using Interactive and Modeling Tools Ainhoa M. Aguado 1 , Andy L. Olivares 1 , Carlos Yagüe 1 , Etelvino Silva 2 , Marta Nuñez-García 1 , Álvaro Fernandez-Quilez 1 , Jordi Mill 1 , Ibai Genua 1 , Dabit Arzamendi 2 , Tom De Potter 3 , Xavier Freixa 4 and Oscar Camara 1* 1 PhySense, Department of Information and Communication Technologies, Universitat Pompeu Fabra, Barcelona, Spain, 2 Division of Interventional Cardiology, Hospital de la Santa Creu i Sant Pau, Universitat Autònoma de Barcelona, Barcelona, Spain, 3 Arrhythmia Unit, Department of Cardiology, Cardiovascular Center, Aalst, Belgium, 4 Department of Cardiology, Hospital Clínic de Barcelona, Universitat de Barcelona, Barcelona, Spain Edited by: According to clinical studies, around one third of patients with atrial fibrillation (AF) will Javier Saiz, suffer a stroke during their lifetime. Between 70 and 90% of these strokes are caused Universitat Politècnica de València, Spain by thrombus formed in the left atrial appendage. In patients with contraindications Reviewed by: to oral anticoagulants, a left atrial appendage occluder (LAAO) is often implanted to Lucy T. Zhang, prevent blood flow entering in the LAA. A limited range of LAAO devices is available, Rensselaer Polytechnic Institute, United States with different designs and sizes. Together with the heterogeneity of LAA morphology, Joakim Sundnes, these factors make LAAO success dependent on clinician’s experience. A sub-optimal Simula Research Laboratory, Norway LAAO implantation can generate thrombi outside the device, eventually leading to *Correspondence: stroke if not treated. The aim of this study was to develop clinician-friendly tools Oscar Camara oscar.camara@upf.edu based on biophysical models to optimize LAAO device therapies. A web-based 3D interactive virtual implantation platform, so-called VIDAA, was created to select the Specialty section: most appropriate LAAO configurations (type of device, size, landing zone) for a given This article was submitted to Computational Physiology and patient-specific LAA morphology. An initial LAAO configuration is proposed in VIDAA, Medicine, automatically computed from LAA shape features (centreline, diameters). The most a section of the journal Frontiers in Physiology promising LAAO settings and LAA geometries were exported from VIDAA to build Received: 16 October 2018 volumetric meshes and run Computational Fluid Dynamics (CFD) simulations to assess Accepted: 22 February 2019 blood flow patterns after implantation. Risk of thrombus formation was estimated from Published: 22 March 2019 the simulated hemodynamics with an index combining information from blood flow Citation: velocity and complexity. The combination of the VIDAA platform with in silico indices Aguado AM, Olivares AL, Yagüe C, Silva E, Nuñez-García M, allowed to identify the LAAO configurations associated to a lower risk of thrombus Fernandez-Quilez Á, Mill J, Genua I, formation; device positioning was key to the creation of regions with turbulent flows after Arzamendi D, De Potter T, Freixa X and Camara O (2019) In silico implantation. Our results demonstrate the potential for optimizing LAAO therapy settings Optimization of Left Atrial Appendage during pre-implant planning based on modeling tools and contribute to reduce the risk Occluder Implantation Using of thrombus formation after treatment. Interactive and Modeling Tools. Front. Physiol. 10:237. Keywords: left atrial appendage occlusion, Computational Fluid Dynamics, in silico optimization of therapies, doi: 10.3389/fphys.2019.00237 web-based implantation platform, atrial fibrillation Frontiers in Physiology | www.frontiersin.org 1 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO 1. INTRODUCTION achieving a complete clinical closure, allows the procedure to be finished earlier since a second device is not required. Atrial Fibrillation (AF) is the most common cardiac arrhythmia A minimal leakage of jet blood flow entering into the LAA diagnosed in clinical practice, affecting around 0.4–1% of the after device implantation is the standard criteria to define a general population (Singh and Holmes, 2010). The atria beat successful closure (López-Mínguez et al., 2014). Unfortunately, irregularly during AF, which leads to ineffective blood flow LAA shape parameters are usually estimated from medical pumping to the ventricles. The consequences of AF have been images with manual tools, being difficult to standardize criteria clearly established in multiple studies, being one of them the to objectively define them. Moreover, values of these parameters increased risk of stroke; one out of three ischemic strokes is coming from different imaging modalities differ substantially related to AF (Boston Scientific, 2018). Between 70 and 90% due to their respective spatial resolution and limitations of AF-related strokes originate from thrombus formed in a (López-Mínguez et al., 2014). cavity of the left atrium, the left atrial appendage (LAA) (Wang Beyond their morphology, studying LA and LAA et al., 2010; Lee et al., 2015). The LAA is a contractile reservoir hemodynamics is also important before and after LAAO and a decompression chamber, depending on the cardiac cycle treatment. The Virchow’s triad defines blood hypercoagulability, phase (Fatkin et al., 1994; Al-Saady et al., 1999). It acts as hemodynamics changes and endothelial wall injury dysfunction a suction during ventricular systole and during diastole as a as the three main players in thrombus formation (Chung and conduit. In addition, the LAA is characterized by its high Lip, 2003). Left atrial appendage blood flow velocities below morphological variability, in parameters such as size, height, some threshold (<0.2 m/s) have been associated to the presence smoothness or number of lobes. The different LAA morphologies of thrombus in the LAA (Beigel et al., 2014). Unfortunately, most are often classified into four categories: chicken wing, cauliflower, clinical studies (Beigel et al., 2014) involving the LAA, usually windsock and cactus (Di Biase et al., 2012). Left atrial motion based on echocardiographic images, only report a single blood irregularity due to AF causes a deceleration of blood flow flow velocity value in one point in space (LAA ostium) and in and stasis, increasing the risk of thrombus formation in the time (end diastole), which constitutes an over-simplification LAA (Markl et al., 2016). The thrombus formed can then of the complex hemodynamics in the LA and LAA. Detailed travel through the circulatory system to the brain, causing a hemodynamics characterization is becoming possible these days cardioembolic stroke. with advanced 4D-flow magnetic resonance imaging technique, Oral anticoagulation (OAC) therapy is usually prescribed on but there are few studies in the LA (Casas et al., 2017; Cibis those AF patients older than 65 years old (Morillo et al., 2017) et al., 2017a,b) and only one in the LAA (Markl et al., 2016). to reduce the risk of thrombus formation. When OAC therapy Some researchers (Zhang and Gay, 2008; Koizumi et al., 2015; is not appropriate, due to high risk of bleeding or because Vedula et al., 2015; Otani et al., 2016; Olivares et al., 2017; Bosi of patient’s limitations, a physical LAA closure is performed et al., 2018; García-Isla et al., 2018) have developed biophysical with the implantation of a left atrial appendage occlusion modeling pipelines for a more detailed in silico analysis of LAA (LAAO) device or surgery-based clipping (Le et al., 2014). blood flow patterns with Computational Fluid Dynamics (CFD) LAAO devices are implanted percutaneously by a femoral vein simulations. However, none of these works studied the effect approach under fluoroscopic and echocardiographic guidance. of implanting a LAAO device with different settings on atrial The most used LAAO devices available in the market (see hemodynamics. In addition, as in the majority of biomedical Figure 1) are the Watchman (Boston Scientific, 2018) and applications, these biophysical models remain an engineering the Amplatzer Amulet (Abbott, 2018), even if new promising tool far from being used to support clinicians on their decisions. devices such as the Lambre (Park et al., 2018) and the Even if LAAO devices are generally becoming an accepted Wavecrest (Saw and Lempereur, 2014) are being proposed option when OAC therapy fails, it still exists controversy every year. In a LAAO procedure, the large intra-subject concerning the safety and the real clinical benefit of LAAO-based anatomical variation of the LAA influences the individual therapy. Although it seems to be effective, short- and long-term implantation. Therefore, biomedical images are acquired to consequences are not fully known (Mandrola et al., 2018). As an have good anatomical knowledge of the cavity, before and example, a recent clinical trial (Fauchier et al., 2018) reported during the intervention. In most clinical centers X-ray and a high probability of getting a device-related thrombus (around transesophageal echocardiography (TEE) images are used to 7%). Moreover, LAAO is recognized as a technically demanding characterize the LAA morphology during the intervention procedure, requiring rigorous training and skills in order to to support decisions on device implantation. Other imaging reduce complications (Tzikas et al., 2016). However, there is a lack modalities such as Computed Tomography (CT) (Chow et al., of advanced computational tools to provide a complete, objective 2017) and 3D Rotational Angiography (3DRA) (De Potter et al., and patient-specific characterization of LAA morphology and 2018) are also being explored in advanced hospitals to have function to assist clinicians on the challenging LAAO device higher-resolution structural information before and during the selection and planning steps. The aim of this work was to intervention, respectively. The ostium (interface betwen LA and develop a clinician-friendly visualization tool to be combined LAA) dimensions and height/depth of the LAA cavity (Morillo with biophysical modeling for LAAO therapy optimization. A et al., 2017) are critical LAA shape parameters to individualize the web-based 3D interactive virtual implantation platform, so- size of the implanted device and the landing zone (location where called VIDAA (Virtual Implantation and Device selection in the device will be released). Appropriate sizing of the device, left Atrial Appendages), was designed and implemented to Frontiers in Physiology | www.frontiersin.org 2 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO TABLE 1 | Patient and device information. Patient Medical image CHAD2VASC Device type Device size (mm) 1 3DRA 3 Watchman 27 2 3DRA 3 Watchman 30 3 3DRA 6 Watchman 27 4 CT 5 Amplatzer Amulet 22 3DRA, 3D Rotational Angiography; CT, Computed Tomography; CHAD2VASC, clinical index characterizing the risk of thrombus formation. LAA and device setting. The VIDAA platform in combination with CFD simulations then allowed the clinician to decide, prior to the intervention, which device configuration was safer, in FIGURE 1 | Pictures (top) and Computational-Aided Design models (bottom) of left atrial appendage occlusion devices. (A) Amplatzer Amulet (Abbott, terms of thrombus formation. 2018); (B) Watchman (Boston Scientific, 2018). 2.1. Patient-Specific Imaging Data and Processing jointly visualize the LA anatomy of an individual patient and Three-dimensional medical images were available for the four different possible LAAO configurations. The VIDAA platform AF patients analyzed in this study: 3DRA images of three initially proposes LAAO setting parameters estimated after a patients from OLV Hospital (Aalst, Belgium); and a CT image morphological analysis of the LAA under study. The clinician of one patient from Hospital Clínic de Barcelona (Spain). later can interactively vary device parameters in VIDAA and The OLV Hospital and Hospital Clínic de Barcelona ethical export outside the platform the most promising configurations. committees approved this study and written informed consent Computational Fluid Dynamics (CFD) simulations were then was obtained from every participant before the procedure. run with a commercial solver for several LAAO configurations Watchman devices were implanted to patients at OLV Hospital extracted from VIDAA to estimate their corresponding risk while an Amplatzer Amulet was the choice at Hospital Clínic de of thrombus formation after implantation. As a first proof of Barcelona (see Table 1). All patients had persistent non-valvular concept, the VIDAA platform was tested on retrospective data atrial fibrillation with high risk of thrombus formation (e.g., of four non-valvular AF patients that underwent a LAA closure. CHAD2VASC > 2). An InnovaTM monoplane fluoroscopy system (GE Healthcare, 2. MATERIALS AND METHODS LLC, Waukesha,WI) was used at OLV Hospital to acquire the 3DRA images, with 0.23 or 0.45 mm pixel size for 512 or 256 The computational workflow designed in this work (see Figure 2) isotropic voxels, respectively. Before and during the injection, is a planning tool that, together with computational modeling, the right ventricle was burst paced at a cycle length of 200 provides enhanced individualized information to the clinician ms during end-expiratory apnea to interrupt the blood flow before the intervention. In all analyzed cases, surface meshes through the LA. All datasets were segmented by an experienced of the LA were reconstructed from binary masks obtained operator on the Advantage Workstation (GE Healthcare) using segmenting high-resolution 3D medical images (CT and 3DRA). the Innova 3D software to reconstruct the LA, which is based In the next step, morphological indices characterizing patient- on threshold and region-growing semi-automatic segmentation specific LAA shapes were automatically calculated. Based on techniques. The CT image at Hospital Clínic de Barcelona was these LAA shape-based indices, a range of optimal LAAO acquired using a SiemensTM machine, synchronized with ECG device parameters (size, position) were estimated. The optimal and supplying 80 ml of contrast, having a volumetric pixel size configuration for the two most popular devices (e.g., Watchman of 0.33 × 0.33 × 0.40 mm for an in-plane matrix of 512 pixels and Amplatzer Amulet) could then interactively be explored and 430 slices. The ITK-Snap1 software (Yushkevich et al., 2006) together with a 3D rendering of the LA in the web-based was used to semi-automatically segment the LA in the CT image VIDAA platform. The clinician could manually modify several (also with region growing and threshold-based techniques, plus LAAO parameters to better understand the case under study and manual corrections). The LA binary masks from segmenting plan the intervention. Once a given LAAO configuration was 3DRA and CT images were the basis to construct surface meshes considered appropriate by the clinician, the VIDAA platform with the Marching Cubes method for subsequent morphological exported all data (e.g., meshes and their spatial relationship) analysis and simulations. A 2D plane was manually selected by that will define the geometrical domain where CFD simulations the operator to define the ostium and the LAA in the LA 3D were run outside VIDAA with a commercial software. Simulation surface mesh. results were post-processed to derive in silico hemodynamics indices estimating the risk of thrombus formation for a given 1 www.itksnap.org Frontiers in Physiology | www.frontiersin.org 3 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 2 | Computational workflow to recommend optimal left atrial appendage occlusion (LAAO) settings for a selected patient. The first stage produces LAA morphological measures from LA segmentations obtained on three-dimensional medical images. Computational 3D renderings of LA/LAA geometries are then jointly visualized with morphological measures and Computer-Aided Design models of LAAO devices in the VIDAA platform. The VIDAA platform allows an interactive and individualized virtual implantation of LAAO devices, generating multiple configurations that can be exported for further processing. Computational-Fluid Dynamics (CFD) simulations run on the exported LAAO configurations provide hemodynamics descriptors to estimate the risk of thrombus formation for each case, which are finally reported to clinicians to support their decisions. 2.1.1. Left Atrial Appendage Morphological simulation to the LAA volumetric mesh. Once the centreline Parameters and Optimal Device Configuration was estimated, several transversal 2D planes, perpendicular to its Two types of morphological parameters were automatically main direction, were selected along its length to compute their estimated from LAA surface meshes: ostium- and centreline- maximal (D1) and minimal (D2) diameters. based parameters. The ostium area, perimeter and principal axes The centreline-based parameters were used to automatically were computed directly from surface mesh elements labeled estimate the optimal LAAO configuration for the analyzed as ostium after the segmentation. The LAA centreline (see anatomy, as illustrated in Figure 3. Following criteria defined by and example in Figure 3) was obtained following a marching our experienced clinicians, device (center) positioning was set algorithm (Genua et al., 2019), where its starting point (the at a geodesic distance along the centreline of 10 and 13 mm furthest point of the LAA) and the centreline direction to the from the ostium for the Amplatzer Amulet and the Watchman, ostium were computed applying a heat transfer propagation respectively. The diameters (D1, D2) of the transversal 2D plane Frontiers in Physiology | www.frontiersin.org 4 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO centered at the optimal device positioning were then computed. The initial orientation of the LAAO device was set perpendicular to the centreline. Based on empirical algorithms from clinicians and official device instructions, the most appropriate LAAO device sizes (three for the Amplatzer Amulet and two for the Watchman) were initially selected. For instance, optimal Amplatzer Amulet device sizes were established following the current clinical protocol at Hospital Clínic de Barcelona: 2–5 or 3–6 mm are usually added to the average mean diameter (between D1 and D2) if imaging data comes from CT or TEE, respectively. If the obtained size is smaller than D1, additional 2–5 mm are added. Finally, only devices ensuring at least a 10% of compression of original size after implantation were selected. Fulfilling the compression-based criteria increments the likelihood of getting the LAAO device attached to the wall for a successful closure, without creating too high stresses on the LAA. The optimal sizes for the Watchman device (see Table A.1 in Appendix 1) were selected based on D1, following official device instructions. 2.2. VIDAA: Virtual Implantation and Device FIGURE 3 | Computational pipeline to obtain optimal left atrial appendage Selection in Left Atrial Appendages occlusion (LAAO) device configuration. Device positioning (e.g., landing zone) VIDAA is a web-based 3D interactive virtual implantation is initially estimated at a certain geodesic distance from the ostium following platform that allows clinicians to select the most appropriate the LAA centreline (10 and 13 mm for the Amplatzer Amulet and Watchman, respectively). Device orientation is set perpendicular to the LAA centreline. LAAO configurations for a given patient. The main features LAAO size is based on the maximum diameters (D1 & D2) of a transversal 2D of the VIDAA platform include: the joint 3D visualization of plane perpendicular to the centreline placed in the optimal positioning. The the LA/LAA surface meshes and the device models; different proposed LAAO devices are initially shown in the VIDAA platform with these modes of mesh visualization and editing tools such as wireframe settings (VIDAA-init), where the clinician can interactively manipulate all their views or clipping, among others; transform controllers for user parameters for a better fitting to the studied LAA morphology (VIDAA-end). manipulation of the device such as translation, rotation and scaling; visualization of LAA morphological parameters such as the centreline, 2D transveral planes of the centreline and its diameters (D1, D2); visualization of the optimal LAA settings to the LA mesh using the Autodesk Meshmixer software2 according to VIDAA such as device size and landing zone; so that simulated flow was developed normal to the surface manual modification of all LAAO settings; and export of a LAAO of the vessels. Volumetric meshes were built with a 3D configuration for subsequent in silico hemodynamics study. Delaunay refinement algorithm available in the Gmsh software Figure 2 illustrates the different steps in the VIDAA workflow. A (default parameters) (Geuzaine and Remacle, 2009)3 for the demonstration video of a proof-of-concept prototype of VIDAA subsequent CFD simulations, resulting in the following number is also available as Supplementary Material (Appendix 2). of tetrahedral elements for the four LA under study: Patient Being a web-based platform, VIDAA is not linked to 1 = 597,550; Patient 2 = 961,015; Patient 3 = 330,091; Patient any software installation, operative system or browser, which 4 = 200,951. The whole meshing pipeline was performed facilitates its usage in a clinical environment. VIDAA was built with the following Open-Source softwares: Gmsh, Autodesk using HTML5 programming language (JavaScript, scss), using MeshMixer, MeshLab4 and Visual Computing Lab5 . The reader different JavaScript libraries such as THREE.js and React.js. The is referred to García-Isla et al. (2018) for more details of the first library, THREE.js, was used to visualize and interact with 3D meshing pipeline. objects (e.g., LA mesh) while React.js was used to separate into The studied LAAO devices (see Figure 1) were built components the different modalities of the platform. using a Computational-Aided Design (CAD) software6 and following the on-line product manual and instructions from the companies (Abbott, 2018; Boston Scientific, 2018). Subsequently, 2.3. Hemodynamics Simulations several LAAO device configurations, including different size and 2.3.1. Mesh Processing Pipeline positioning settings, were exported from the VIDAA platform. To ensure convergence of hemodynamics simulations, LA meshes from medical image segmentations required several pre- 2 http://www.meshmixer.com/ processing phases (see Figure 4). First, a Taubin smoothing 3 http://gmsh.info/ filter (with scale factors λ = 0.6, µ = –0.53 and 10 iterations) 4 http://www.meshlab.net/ was applied. Subsequently, cylinders representing the pulmonary 5 http://vcg.isti.cnr.it/ veins (PV) and the mitral valve (MV) were manually inserted 6 Geometry-Design Modeler (Ansys 18.2) Frontiers in Physiology | www.frontiersin.org 5 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 4 | Mesh processing pipeline for hemodynamics simulations: (A) Computed Tomography (CT) image; (B) Surface mesh after CT segmentation; (C) Removal of structures surrounding the left atrium (LA); (D) Insertion of cylinders for the pulmonary veins (PV); (E) Volumetric LA mesh with two right pulmonary veins (RPVs), two left pulmonary veins (LPVs), a mitral valve (MV) and the left atrial appendage (LAA). Volumetric meshes for all configurations under study were in the LAA of the ECAP index for different simulation scenarios obtained defining two surfaces in Gmsh: one for the LA and was analyzed with histograms showing the percentage of nodes another for the implanted device. in different ECAP intervals. 2.3.2. Computational Fluid Dynamics 2.3.4. Simulation Scenarios The CFD analysis was performed using ANSYS Fluent 18.2, in In silico blood flow patterns were studied in different simulation which blood was modeled as an incompressible Newtonian fluid scenarios to study the influence of the implanted device (density of ρ= 1,060 kg/m3 ; dynamic viscosity of µ = 0.0035 Pa·s). configuration. Four simulation scenarios were studied for each Blood was simulated using the incompressible Navier-Stokes patient, resulting in a total number of 16 simulations: and continuity equations. Model dynamics was introduced with • Without LAAO device (No LAAO). a sinusoidal function representing a heart cycle of a healthy • LAAO device with size and positioning derived from patient (Fernandez-Perez et al., 2012; García-Isla et al., 2018). morphological analysis and criteria defined by experienced Boundary conditions at systole (first 0.40 s) were modeled with clinicians (see section 2.1.1); this is the default configuration the PVs as velocity-inlets and the MV as a wall, to represent that shown in the VIDAA platform (VIDAA-init). the valve is closed during this cardiac phase (Fernandez-Perez • LAAO device with VIDAA-init settings but with a 20% et al., 2012). The opening of the MV in diastole (duration of 0.65 smaller size. s) was modeled applying an outlet pressure of 8 mmHg (Nagueh • LAAO device VIDAA-init settings but with a sub-optimal et al., 2008) to the MV. All LA walls were modeled as rigid walls position and orientation (Misplaced). with no slip, representing the worst scenario in (persistent) AF, • LAAO device after manipulation in VIDAA aiming where the atrium barely contracts anymore. The LAAO device at uniformly covering the union between the LA and was also modeled as a wall to represent that no blood could go LAA (VIDAA-end). through it. 2.3.3. In silico Indices for Risk of Thrombus Formation 3. RESULTS As in García-Isla et al. (2018), we estimated several in silico indices to characterize blood flow patterns derived from 3.1. Blood Flow Velocity Patterns CFD simulations and identify pro-thrombotic regions in the Streamlines and velocity vector fields characterizing blood flow LAA, usually where blood flow is complex and velocities are patterns in the LA/LAA for different simulation scenarios were low (Achille et al., 2014). For the sake of simplicity, only the analyzed. Figure 5 shows simulation results for the four patients Endothelial Cell Activation Potential (ECAP) index is reported under study and the LAAO configurations described above, in here since it combines the Time-Averaged Wall Shear Stress diastole (t = 0.55 s, ventricular filling), when blood flow velocities (TAWSS) with the Oscillatory Shear Index (OSI): low velocity in the LAA are higher. Videos of the hemodynamics simulations (small TAWSS values) and high complex (large OSI values) blood (blood flow velocity streamlines for the whole cardiac cycle) for flows will produce high ECAP values, thus indicating a region all patients without a device and with the VIDAA-end settings are with a high risk of thrombus formation. included as Supplementary Material (see Appendix 2). The equations to derive these indices are included in At baseline (without an implanted LAAO device, No LAAO Appendix 3. For a qualitative analysis, the in silico indices were configuration, first row in Figure 5), blood flow is entering the visualized (using ParaView 5.2.0) as colored maps superimposed LAA in all patients with relatively low velocities, in some cases on the 3D LA geometry, providing an intuitive way to detect with complex patterns (e.g., Patient 3) in secondary lobes. The regions with irregular blood flow. Furthermore, the distribution device configuration proposed by the VIDAA platform (both Frontiers in Physiology | www.frontiersin.org 6 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 5 | Blood flow streamlines in the left atrial appendage (LAA) from hemodynamics simulations of the four patients under study. Snapshots capture the flow behavior in diastole, when higher velocities are present (t = 0.55 s). No LAAO: without device; VIDAA-init: device settings from morphological analysis and clinical recommendations; VIDAA-end: device settings after interactive manipulation in VIDAA; 20% smaller: device with a 20% smaller size than VIDAA-init; Misplaced: device with sub-optimal position and orientation. VIDAA-init and VIDAA-end) achieved a perfect occlusion in the device, allowing more flow through the LAA (fourth row, Patient 1 (second and third row, first column, in Figure 5). second column, in Figure 5). On the other hand, Patient 3 On the other hand, blood flow was penetrating the LAA in (third column in Figure 5) showed a correct flow blockage in the same case when decreasing the LAAO device size (20% all device configurations. Finally, Patient 4 showed fast velocities Smaller scenario), showing a potential stagnation process, with and laminar flow in the LA cavity outside the implanted device low velocities, in the lateral part of the occluder (blue arrows for the VIDAA-end configuration (red streamlines in third row, next to the device; fourth row, first column, in Figure 5). All fourth column in Figure 5), which was not achieved with the configurations have the device positioning deeper in the LAA remaining settings. than the VIDAA-end, creating a non-closed cavity between the Figure 6 depicts blood flow velocity curves over the whole ostium and the device where new complex blood flow loops cardiac cycle in a point outside the device (following the LAA appeared. Patient 2 presented a similar behavior to Patient 1 centerline direction toward the main LA cavity) for all simulation but with a more visible negative effect of reducing the size of scenarios in the four patients analyzed in this work. We can Frontiers in Physiology | www.frontiersin.org 7 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 6 | Blood flow velocity curves over the whole cardiac cycle in a point outside the device in the left atrial cavity for all simulation scenarios in the four patients. VIDAA-init: device settings from morphological analysis and clinical recommendations; VIDAA-end: device settings after interactive manipulation in VIDAA; 20% smaller: device with a 20% smaller size than VIDAA-init; Misplaced: device with sub-optimal position and orientation. appreciate that the VIDAA-end configuration (red curve in located in several areas of the LAA depending on morphology Figure 6) leads to velocity profiles with higher magnitudes in complexity. For instance, Patient 4 is the case showing smaller diastole for some cases (e.g., see Patient 4), suggesting faster flow ECAP values due to its smooth morphology and orientation circulation and thus less risk of thrombus formation due to device with respect to the main LA cavity, leading to larger blood implantation. On the other hand, results are inconclusive for flow velocity magnitudes in most regions (except in its tip). other patients (e.g., see Patient 3), exemplifying the difficulty of More importantly, high ECAP areas appear more prominently characterizing the 4D nature of blood flow patterns by observing in device configurations with incorrect LAAO settings (e.g., 1D velocity profiles in an individual spatial point. 20% smaller), whereas the VIDAA-end configuration seems to minimize the presence of these areas. 3.2. In silico Indices for Risk of Thrombus Figure 8 plots the histograms of ECAP intervals as Formation percentages with respect to surface area for each patient Figure 7 shows the ECAP distribution mapped as a colormap and different LAAO scenarios. The ECAP maps were analyzed onto the LA/LAA geometries for all simulation scenarios in the within the region between the device and the LAA ostium. For patients under study. For LAAO scenarios (i.e., all except the the four analyzed LAA geometries, the VIDAA-end settings No LAAO), the ECAP map is only displayed between the LAA resulted in the lower percentage of surface areas with higher ostium and the device to study the risk of thrombus formation ECAP values (>3). We can also observe that the VIDAA-init outside the device. It can easily be observed at baseline (No settings were not always adequate, sometimes even performing LAAO configuration; first row in Figure 7) that higher values worse than the Misplaced and 20% smaller scenarios, illustrating of ECAP (i.e., complex flows and low velocities; red-green areas the benefit of interactively manipulating the device settings in in Figure 7), thus with a higher risk of thrombus formation, are the VIDAA platform. Frontiers in Physiology | www.frontiersin.org 8 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 7 | Endothelial Cell Activation Potential (ECAP) distribution (Pa−1 ) visualized as colored maps superimposed to the LA wall for the four patients under study. No LAAO: without device; VIDAA-init: device settings from morphological analysis and clinical recommendations; VIDAA-end: device settings after interactive manipulation in VIDAA; 20% smaller: device with a 20% smaller size than VIDAA-init; Misplaced: device with sub-optimal position and orientation. 4. DISCUSSION currently possible with existing imaging modalities. Providing such information for different LAAO settings to the clinician In this study, different configurations of LAA occlusion devices would be quite beneficial for a better and faster planning of were evaluated in four patient-specific left atria using advanced the intervention. computational tools. A web-based platform, so-called VIDAA, The implantation of LAAO devices is usually quite successful was developed in which LAAO devices could be virtually (around 95%, defined as small flow jets into the LAA implanted, interactively selecting the desired configuration for a after implantation), with peri-procedural complicate rates of given individual. Once a device configuration was chosen, it was approximately 5% (Grosset-Janin et al., 2015). Nevertheless, exported out of the platform to built a computational domain residual bleedings are not uncommon and the reduction of where hemodynamics CFD-based simulations were run. The thrombus formation risk, even if superior to OAC-based resulting blood flow simulations were post-processed to generate therapy, could be improved (60–75% Grosset-Janin et al., quantitative in silico indices (e.g., ECAP) to locally assess the risk 2015). Moreover, it is not well-known when patients can stop of thrombus formation after device implantation, which is not anti-coagulants after LAAO implantation, which can lead to Frontiers in Physiology | www.frontiersin.org 9 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO FIGURE 8 | Histograms of Endothelial Cell Activation Potential (ECAP) distribution (Pa−1 ) as a percentage of surface area for each patient and different device scenarios. VIDAA-init: device settings from morphological analysis and clinical recommendations; VIDAA-end: device settings after interactive manipulation in VIDAA; 20% smaller: device with a 20% smaller size than VIDAA-init; Misplaced: device with sub-optimal position and orientation. device-related thrombus formation (7% of cases in a recent analysis and clinical recommendations (VIDAA-init), repeatedly study Fauchier et al., 2018). The development of advanced created complex flow loops with low velocities around the device, computational tools including 3D medical imaging, clinician- incrementing the risk of thrombus formation post-intervention. friendly interfaces and biophysical modeling to fully characterize However, the VIDAA-end positioning of the device may not be the LAA 3D morphology, 4D blood flow patterns and virtually adequate if it leads to a blockage of the circumflex artery, which predict the individual risk of thrombus formation for different is located just outside the LAA ostium. Unfortunately, we did not device settings, would contribute to the acceptance of LAAO have information about the circumflex artery exact location in therapies, properly assessing their cost-effectiveness and clinical the studied cases. In the future, this information will be obtained benefit as well identifying their limitations, which is still under from CT images and added as an additional constraint in the debate (Hu and Yogeswaran, 2015; Mandrola et al., 2018). VIDAA platform. The obtained results confirm the relevance of an appropriate Device sizing and positioning are then crucial, assuming LAAO device configuration, personalized to each patient, to a good attachment to the LAA wall. Nowadays, the process ensure a complete occlusion of the LAA cavity and minimize for selecting the LAAO size and defining the landing zone the creation of areas around the implanted device prone to where the device will be released is too subjective and prone thrombus formation (e.g., low velocities and complex flow). to errors due to the lack of quantitative techniques and high- Consistently, LAAO settings obtained after device manipulation resolution 3D images. We have observed in our study that in the VIDAA platform (VIDAA-end) were the most appropriate, small changes in the selection of transversal 2D planes along ensuring a complete blockage of the LAA cavity in all cases. the LAA centerline produced maximum diameters in the range On the contrary, other LAAO configurations, including the of 22–31 mm in the same geometry. Similar variability can VIDAA initial configuration based on LAA morphological be found when using 2D-based or different medical imaging Frontiers in Physiology | www.frontiersin.org 10 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO modalities, as it was demonstrated in López-Mínguez et al. (between device and LA wall) or discrete particle models (2014), where LAAO dimensions were consistent from X- (for coagulation and thrombus formation Hathcock, 2006) ray, echocardiography and CT images in only 23% of the may provide more realistic simulations, at the expense of cases. Therefore, a full 3D characterization of the LA/LAA increased computational costs and complexity. In addition, geometry (i.e., dimensions, relative orientation) is required patient-specific blood flow boundary conditions (e.g., based for an optimal device sizing/positioning and avoid generating on Doppler data), rather than the generic curves used in areas between the implanted device and the atrial wall prone this study, would be required for an individualized assessment to thrombus formation. A reduced learning curve of the of hemodynamics. LAAO device implantation process and more comprehensive The post-processing of CFD simulations has been one of intervention planning has already been demonstrated by using the most critical steps in the developed modeling pipeline. CT images (Wang et al., 2016; Chow et al., 2017). Recently (De For instance, the visualization of blood flow streamlines is Potter et al., 2018), 3DRXA has also been proven an interesting highly dependent on the location of the seeds required to alternative to echocardiographic images during the intervention. generate them. An incorrect seed placement can lead to The VIDAA platform has then been developed to allow misinterpretation of CFD simulations and wrong conclusions. the interactive 3D visualization of LA/LAA geometries and To study LAA blood flow patterns in the LAA it is CAD models of the devices. The integration of VIDAA with convenient to place seeds near the LAA ostium, according hemodynamics simulations has been very helpful to identify to our experience. Additionally, one needs to be cautious areas with high risk of thrombus formation under certain on using velocity 1D profiles and values evaluated in a LAAO configurations. The ECAP parameter, initially proposed single point in space as an estimation of blood flow pattern by Achille et al. (2014) in an aortic aneurysm application, behavior. It is not obvious to select a point in which reliably provided consistent results, with higher values in the LAA comparing velocity profiles for different device configurations, and showing its potential to estimate thrombogenesis. Overall, as shown in Figure 6. This is also critical for the validation the VIDAA platform and biophysical models should provide a of simulation results with clinical data. We have performed more complete way to analyze the relation between LAA shape, initial visual comparisons between mitral valve and LAA velocity hemodynamics, device settings and thrombus formation to the profiles from our simulations and from Doppler data of the clinicians prior to the intervention. studied patients, showing promising similarities. Nevertheless, The presented methodology has several limitations that a more exhaustive and quantitative validation with more could be improved in the future. For instance, the 3D complete Doppler or 4D-flow MRI data, ideally on several model construction, including image processing steps (e.g., geometries and with different LAAO devices, is still missing. segmentation, definition of ostium plane) and the meshing In addition, the analysis of a larger number of cases would pipeline, is semi-automatic, which could lead to intra-operator allow identifying patients with thromboembolic history and inconsistencies. The obtained image processing results and compare their morphological and hemodynamics parameters corresponding 3D models were qualitatively evaluated by several given by VIDAA and biophysical models with patients without observers for each case. However, even if developing a fully a thrombus. automatic image processing pipeline was out of the scope of this paper, left atrial segmentation techniques based on deep-learning algorithms are already becoming available (see recent Atrial 5. CONCLUSIONS Challenge at STACOM18 workshop7 ). Sensitivity analysis and convergence studies will also identify the most critical meshing The VIDAA platform was developed to perform virtual pipeline parameters, such as the required mesh resolution, to implantation of LAAO devices on patient-specific geometries obtain robust wall shear stress maps, which is critical to estimate in an interactive way. Its combination with biophysical reliable ECAP maps. models of hemodynamics and complete LAA shape 3D Moreover, several boundary conditions and modeling choices characterization offers clinicians unprecedented computational could be upgraded to achieve more realistic simulations. The tools to select the optimal device settings prior to the assumption of rigid LA walls for a persistent AF patient is intervention. Different LAAO device settings can be tested valid from a clinical point of view. Nevertheless, the LAA to minimize the areas prone to thrombus formation after is quite a flexible structure and the mitral ring displacement device implantation, according to in silico indices derived is non-negligible to properly simulate LA reservoir, conduit from CFD simulation results. The LAAO configurations found and booster pump phases. Motion-based constraints could after manipulation of device settings in the VIDAA platform be imposed from patient-specific information, as in Otani were linked to a reduced risk of thrombus formation outside et al. (2016) from CT images or similar. Moreover, boundary the implanted device, according to a qualitatively analysis of conditions such as the inlet velocity profile should be adapted blood flow streamlines and ECAP maps. Future work will to atrial fibrillation (i.e., absence of atrial contraction). More be focused on the evaluation of the VIDAA platform in a sophisticated numerical techniques such as Fluid-Structure clinical environment, analyzing a larger number of LAA cases interaction (between blood flow and LA wall), contact-based and exploring its use as an alternative to 3D printing for interventional planning, also in combination with immersive 7 http://atriaseg2018.cardiacatlas.org/ visualization interfaces. Frontiers in Physiology | www.frontiersin.org 11 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO DATA AVAILABILITY FUNDING Simulation data including solver configurations and results are This work was supported by the Spanish Ministry of Economy available under request. A beta version of the VIDAA platform and Competitiveness under the Maria de Maeztu Units of is available at the GitHub account of the BCN-Medtech research Excellence Programme (MDM-2015-0502) and the Retos I+D unit (https://github.com/bcnmedtech). Programme (DPI2015-71640-R). AUTHOR CONTRIBUTIONS ACKNOWLEDGMENTS AA generated CFD models, processed and analyzed data, We would like to thank Dr. Brugaletta from Hospital Clínic developed VIDAA platform, and wrote the paper. AO conceived (Barcelona, Spain) for his support and additional information the research and VIDAA platform, designed CFD models and about LAAO-based therapies. We would like to give an analyzed data, wrote the paper. CY developed VIDAA platform, special mention to the undergraduate students from Biomedical gave technical support on the web tools. ES provided clinical data Engineering degree involved in the development of the first CFD from OLV and performed segmentation of 3DRA images. MN-G models (E. Gonzalez, T. Orejon, J. Aviles, I. Sainz, A. Iribar, gave technical support on LA and LAA mesh processing. AF-Q J. Pose and A. Serra). Also thank G. García for providing the computed LAA morphological measures and created database. code to obtain semi-automatically the numerical indices for risk JM gave technical support on reconstruction of CT images of thrombus formation, and M. P. García for her help in the and CFD models. IG developed the algorithm for centreline development of VIDAA platform. extraction. TDP provided data from OLV and practical context. DA provided data from Hospital de la Santa Creu i Sant SUPPLEMENTARY MATERIAL Pau of Barcelona and set specifications for VIDAA platform. XF provided data from Hospital Clínic de Barcelona and set The Supplementary Material for this article can be found specifications for VIDAA platform. OC oversaw the whole online at: https://www.frontiersin.org/articles/10.3389/fphys. project, conceived the research, and wrote-revised the paper. 2019.00237/full#supplementary-material REFERENCES electrical cardioversion of atrial fibrillation. Front. Physiol. 8:1052. doi: 10.3389/fphys.2017.01052 Abbott (2018). AMPLATZERTM Left Atrial Appendage Occluder. Available online De Potter, T., Chatzikyriakou, S., Silva, E., Camp, G., and Penicka, M. at: https://www.sjmglobal.com (2018). A pilot study for left atrial appendage occlusion guided by 3- Achille, P. D., Tellides, G., Figueroa, C. A., and Humphrey, J. D. (2014). A Dimensional rotational angiography alone. JACC Cardiovasc. Interv. 11, 223– haemodynamic predictor of intraluminal thrombus formation in abdominal 224. doi: 10.1016/j.jcin.2017.08.053 aortic aneurysms A haemodynamic predictor of intraluminal thrombus Di Biase, L., Santangeli, P., Anselmino, M., Mohanty, P., Salvetti, I., Gili, S., et al. formation in abdominal aortic aneurysms. Proc. R. Soc. A 470:20140163. (2012). Does the left atrial appendage morphology correlate with the risk of doi: 10.1098/rspa.2014.0163 stroke in patients with atrial fibrillation? J. Am. Coll. Cardiol. 60, 531–538. Al-Saady, N. M., Obel, O. A., and Camm, A. J. (1999). Left atrial appendage: doi: 10.1016/j.jacc.2012.04.032 structure, function, and role in thromboembolism. Heart 82, 547–555. Fatkin, D., Kelly, R. P., and Feneley, M. P. (1994). Relations between left doi: 10.1136/hrt.82.5.547 atrial appendage blood flow velocity, spontaneous echocardiographic contrast Beigel, R., Wunderlich, N. C., Ho, S. Y., Arsanjani, R., and Siegel, R. J. (2014). and thromboembolic risk in vivo. J. Am. Coll. Cardiol. 23, 961–969. The left atrial appendage: anatomy, function, and noninvasive evaluation. JACC doi: 10.1016/0735-1097(94)90644-0 Cardiovasc. Imaging 7, 1251–1265. doi: 10.1016/j.jcmg.2014.08.009 Fauchier, L., Cinaud, A., Brigadeau, F., Lepillier, A., Pierre, B., Abbey, Bosi, G. M., Cook, A., Rai, R., Menezes, L. J., Schievano, S., Torii, R., et al. (2018). S., et al. (2018). Device-related thrombosis after percutaneous left atrial Computational fluid dynamic analysis of the left atrial appendage to predict appendage occlusion for atrial fibrillation. J. Am. Coll. Cardiol. 71, 1528–1536. thrombosis risk. Front. Cardiovas. Med. 5:34. doi: 10.3389/fcvm.2018.00034 doi: 10.1016/j.jacc.2018.01.076 Boston Scientific (2018). WATCHMAN Left Atrial Appendage Closure Device. Fernandez-Perez, G. C., Duarte, R., Corral de la Calle, M., Calatayud, J., Available online at: https://www.watchman.com and Sanchez Gonzalez, J. (2012). Analysis of left ventricular diastolic Casas, B., Lantz, J., Viola, F., Cedersund, G., Bolger, A. F., Carlhäll, function using magnetic resonance imaging. Radiologia 54, 295–305. C. J., et al. (2017). Bridging the gap between measurements and doi: 10.1016/j.rxeng.2011.09.003 modelling: a cardiovascular functional avatar. Sci. Rep. 7, 1–15. García-Isla, G., Olivares, A. L., Silva, E., Nuñez-Garcia, M., Butakoff, C., Sanchez- doi: 10.1038/s41598-017-06339-0 Quintana, D., et al. (2018). Sensitivity analysis of geometrical parameters to Chow, D. H., Bieliauskas, G., Sawaya, F. J., Millan-Iturbe, O., Kofoed, K. F., study haemodynamics and thrombus formation in the left atrial appendage. Søndergaard, L., et al. (2017). A comparative study of different imaging Int. J. Numer. Methods Biomed. Eng. 34, 1–14. doi: 10.1002/cnm.3100 modalities for successful percutaneous left atrial appendage closure. Open Genua I., Olivares, A. L., Silva, E., Mill, J., Fernández, Á., Aguado, A., et al. (2019). Heart 4:e000627. doi: 10.1136/openhrt-2017-000627 “Centreline-based shape descriptors of the left atrial appendage in relation Chung, I., and Lip, G. Y. (2003). Virchow’s triad revisited: blood constituents. with thrombus formation,” in Statistical Atlases and Computational Models of Pathophysiol. Haemost. Thromb. 33, 449–454. doi: 10.1159/000083844 the Heart. Atrial Segmentation and LV Quantification Challenges, STACOM Cibis, M., Bustamante, M., Eriksson, J., Carlhäll, C. J., and Ebbers, T. (2017a). 2018, Lecture Notes in Computer Science, Vol. 11395 (Granada), 200–208. Creating hemodynamic atlases of cardiac 4D flow MRI. J. Magn. Res. Imaging doi: 10.1007/978-3-030-12029-0_22 46, 1389–1399. doi: 10.1002/jmri.25691 Geuzaine, C., and Remacle, J.-F. (2009). Gmsh: a 3-d finite element mesh generator Cibis, M., Lindahl, T. L., Ebbers, T., Karlsson, L. O., and Carlhäll, C. J. with built-in pre- and post-processing facilities. Int. J. Numer. Methods Eng. 79, (2017b). Left atrial 4D blood flow dynamics and hemostasis following 1309–1331. doi: 10.1002/nme.2579 Frontiers in Physiology | www.frontiersin.org 12 March 2019 | Volume 10 | Article 237 Aguado et al. Optimization of LAAO Grosset-Janin, D., Barth, E., Bertrand, B., and Detante, O. (2015). Percutaneous Park, J. W., Sievert, H., Kleinecke, C., Vaskelyte, L., Schnupp, S., Sievert, left atrial appendage occlusion for stroke prevention in patients with atrial K., et al. (2018). Left atrial appendage occlusion with lambre in atrial fibrillation and contraindication for anticoagulation. Rev. Neurol. (Paris) 171, fibrillation: initial European experience. Int. J. Cardiol. 265, 97–102. 426–432. doi: 10.1016/j.neurol.2014.11.009 doi: 10.1016/j.ijcard.2018.02.120 Hathcock, J. J. (2006). Flow effects on coagulation and thrombosis. Arterioscler. Saw, J., and Lempereur, M. (2014). Percutaneous left atrial appendage closure Thromb. Vasc. Biol. 26, 1729–1737. doi: 10.1161/01.ATV.0000229658.76797.30 procedural techniques and outcomes. JACC Cardiovasc Interv. 7, 1205–1220. Hu, T. Y., and Yogeswaran, V. (2015). Device-based approach to prevention doi: 10.1016/j.jcin.2014.05.026 of stroke in atrial fibrillation. Innov. Cardiac Rhythm Manag. 6, 2038–2050. Singh, I. M., and Holmes, D. R. (2010). Left atrial appendage closure. Curr. Cardiol. doi: 10.19102/icrm.2015.060603 Rep. 12, 413–421. doi: 10.1007/s11886-010-0122-9 Koizumi, R., Funamoto, K., Hayase, T., Kanke, Y., Shibata, M., Shiraishi, Y., et al. Tzikas, A., Gafoor, S., Meerkin, D., Freixa, X., Cruz-Gonzalez, I., Lewalter, T., et al. (2015). Numerical analysis of hemodynamic changes in the left atrium due to (2016). Left atrial appendage occlusion with the AMPLATZER Amulet device: atrial fibrillation. J. Biomech. 48, 472–478. doi: 10.1016/j.jbiomech.2014.12.025 an expert consensus step-by-step approach. EuroIntervention 11, 1512–1521. Le, D. L., Khodjaev, S. D., and Morelli, R. L. (2014). Percutaneous methods of left doi: 10.4244/EIJV11I13A292 atrial appendage exclusion: an alternative to the internist. J. Commun. Hosp. Vedula, V., George, R., Younes, L., and Mittal, R. (2015). Hemodynamics in the Intern. Med. Perspect. 4:0–5. doi: 10.3402/jchimp.v4.22719 left atrium and its effect on ventricular flow patterns. J. Biomech. Eng. 137, 1–8. Lee, J. M., Seo, J., Uhm, J.-S., Kim, Y. J., Lee, H.-J., Kim, J.-Y., et al. (2015). Why doi: 10.1115/1.4031487 is left atrial appendage morphology related to strokes? an analysis of the flow Wang, D. D., Eng, M., Kupsky, D., Myers, E., Forbes, M., Rahman, M., velocity and orifice size of the left atrial appendage. J. Cardiovasc. Electrophysiol. et al. (2016). Application of 3-Dimensional computed tomographic image 26, 922–927. doi: 10.1111/jce.12710 guidance to WATCHMAN implantation and impact onEarly operator learning López-Mínguez, J., González-Fernández, R., Fernández-Vegas, C., Millán-Nuñez, curve: single-center experience. JACC Cardiovasc. Interv. 9, 2329–2340. V., Eugenia Fuentes-Cañamero, M., Manuel Nogales-Asensio, J., et al. doi: 10.1016/j.jcin.2016.07.038 (2014). Comparison of imaging techniques to assess appendage anatomy and Wang, Y., Di Biase, L., Horton, R. p., Nguyen, T., Morhanty, P., and Natale, measurements for left atrial appendage closure device selection. J. Invasive A. (2010). Left atrial appendage studied by computed tomography to help Cardiol. 26, 462–467. planning for appendage closure device placement. J. Cardiovasc. Electrophysiol. Mandrola, J., Foy, A., and Naccarelli, G. (2018). Percutaneous left atrial appendage 21, 973–982. doi: 10.1111/j.1540-8167.2010.01814.x closure is not ready for routine clinical use. Heart Rhythm. 15, 298–301. Yushkevich, P. A., Piven, J., Hazlett, H. C., Smith, R. G., Ho, S., Gee, J. C., et al. doi: 10.1016/j.hrthm.2017.10.007 (2006). User-guided 3D active contour segmentation of anatomical structures: Markl, M., Lee, D. C., Furiasse, N., Carr, M., Foucar, C., Ng, J., et al. (2016). Significantly improved efficiency and reliability. NeuroImage 31, 1116–1128. Atrial structure and function left atrial and left atrial appendage 4D blood doi: 10.1016/j.neuroimage.2006.01.015 flow dynamics in atrial fibrillation. Circ. Cardiovasc. Imaging 9:e004984. Zhang, L. T., and Gay, M. (2008). Characterizing left atrial appendage doi: 10.1161/CIRCIMAGING.116.004984 functions in sinus rhythm and atrial fibrillation using computational Morillo, C. A., Banerjee, A., Perel, P., Wood, D., and Jouven, X. (2017). models. J. Biomech. 41, 2515–2523. doi: 10.1016/j.jbiomech.2008. Atrial fibrillation: the current epidemic. J. Geriatr. Cardiol. 14, 195–203. 05.012 doi: 10.11909/j.issn.1671-5411.2017.03.011 Nagueh, S. F., Appleton, C. P., Gillebert, T. C., Marino, P. N., Oh, J. K., Smiseth, Conflict of Interest Statement: The authors declare that the research was O. A., et al. (2008). Recommendations for the evaluation of left ventricular conducted in the absence of any commercial or financial relationships that could diastolic function by echocardiography. Eur. J. Echocardiogr. 10, 165–193. be construed as a potential conflict of interest. doi: 10.1093/ejechocard/jep007 Olivares, A. L., Silva, E., Nuñez-Garcia, M., Butakoff, C., Sánchez-quintana, D., Copyright © 2019 Aguado, Olivares, Yagüe, Silva, Nuñez-García, Fernandez-Quilez, Freixa, X., et al. (2017). Functional Imaging and Modelling of the Heart, Vol. Mill, Genua, Arzamendi, De Potter, Freixa and Camara. This is an open-access 10263 of Lecture Notes in Computer Science. Cham: Springer International article distributed under the terms of the Creative Commons Attribution License (CC Publishing. BY). The use, distribution or reproduction in other forums is permitted, provided Otani, T., Al-Issa, A., Pourmorteza, A., McEigh, E., Wada, S., and Ashikaga, the original author(s) and the copyright owner(s) are credited and that the original H. (2016). A computational framework for personalized blood flow publication in this journal is cited, in accordance with accepted academic practice. analysis in the human left atrium. Ann. Biomed. Eng. 44, 3284–3294. No use, distribution or reproduction is permitted which does not comply with these doi: 10.1007/s10439-016-1590-x terms. Frontiers in Physiology | www.frontiersin.org 13 March 2019 | Volume 10 | Article 237