Statistical Atlases and Computational Models of the Heart

As decisions in cardiology increasingly rely on non-invasive methods, fast and precise image analysis tools have become a crucial component of the clinical workflow. Especially when dealing with complex cardiovascular disorders, such as valvular heart disease, advanced imaging techniques have the potential to significantly improve treatment outcome as well as to reduce procedure risks and related costs. We are developing patient-specific cardiac models, estimated from available multi-modal images, to enable advanced clinical applications for the management of cardiovascular disease. In particular, a novel physiological model of the complete heart, including the chambers and valvular apparatus is introduced, which captures a large spectrum of morphological, dynamic and pathological variations. To estimate the patient-specific model parameters from four-dimensional cardiac images, we have developed a robust learning-based framework. The model-driven approach enables a multitude of advanced clinical applications. Gold standard clinical methods, which manually process 2D images, can be replaced with fast, precise, and comprehensive model-based quantification to enhance cardiac analysis. For emerging percutaneous and minimal invasive valve interventions, cardiac surgeons and interventional cardiologists can substantially benefit from automated patient selection and virtual valve implantation techniques. Furthermore, the complete cardiac model enables for patient-specific hemodynamic simulations and blood flow analysis. Extensive experiments demonstrated the potential of these technologies to improve treatment of cardiovascular disease.

A framework for the automatic extraction and generation of patient-specific organ models from different image modalities is presented. These models can be used to extract and represent diagnostic information about the heart and its function. Furthermore, the models can be used for treatment planning and an overlay of the models onto X-ray fluoroscopy images can support navigation when performing an intervention in the CathLab.

The Cardiac Atlas Project (CAP) is a NIH sponsored international collaboration to establish a web-accessible structural and functional atlas of the normal and pathological heart as a resource for the clinical, research and educational communities. An initial goal of the atlas is to facilitate statistical analysis of regional heart shape and wall motion characteristics, and characterization of remodeling, between and within population groups. The two main early contributing studies are the Multi Ethnic Study of Atherosclerosis (MESA) and the Defibrillators to Reduce Risk by Magnetic Resonance Imaging Evaluation (DETERMINE) clinical trial. De-identified image and text data from 2864 asymptomatic volunteers from MESA, and 470 myocardial infarction cases from DETERMINE, are currently available in the CAP database. DICOM images were de-identified using HIPAA compliant software based on tools provided by the Center for Computational Biology at UCLA. Only those cases with informed consent and IRB approval compatible with the CAP were included. Researchers requesting permission to access CAP data can apply through the CAP website (www.cardiacatlas.org). All proposals for data access must be approved by the data contributors, and applicants must sign a data transfer agreement with each study from which data is requested. Software to visualize cardiac images and create 3D mathematical models, developed in the CAP, is available open-source from the website.

The Cardiac Atlas Project seeks to establish a standardized database of cardiovascular imaging examinations, together with derived analyses, for the purposes of statistical characterization of global and regional heart function abnormalities. We present preliminary results from a subset of cases contributed from the Defibrillators to Reduce Risk by Magnetic Resonance Imaging Evaluation (DETERMINE) study of patients with myocardial infarction. Finite element models were fitted to the epicardial and endocardial surfaces throughout the cardiac cycle in 200 patients using a standardized procedure. The control points of the shape model were used in a principal component analysis of shape and motion. The modes were associated with well-known clinical indices of adverse remodeling in heart disease, including heart size, sphericity and mitral valve geometry. These results therefore show promise for the clinical application of a statistical analysis of shape and motion in patients with myocardial infarction.

We describe the software design, architecture and infrastructure employed in the Cardiac Atlas Project (CAP), an international collaboration to establish a web-accessible structural and functional atlas of the normal and pathological heart. Cardiac imaging data is de-identified in a HIPAA compliant manner using the LONI Debabeler with customized DICOM mappings. A production database and web-interface were employed based on existing tools developed by LONI. A new open-source database and web interface have been developed for research purposes. After consideration and evaluation of several software frameworks, the research database has been implemented based on a 3-tier architecture utilizing MySQL, JBoss and Dcm4chee. Parts of Dcm4chee have been extended to enable access to MRI specific attributes and arbitrary search parameters. An XML schema has been designed representing the elements associated with the creation and curation of volumetric shape models. The research database is implemented compliant to the DICOM standard, thus providing compatibility with existing clinical networks and devices. A modeling tool, the CAP client, has been developed to enable browsing of 3D image data and creation and modification of volumetric shape models. All software components developed by the CAP are open source and are freely available under the Mozilla license.

In this paper, we present the use of atlas-based indexes of abnormality for the quantification of cardiac resynchronization therapy (CRT) outcome in terms of motion. We build an atlas of normal motion from 21 healthy volunteers to which we compare 88 CRT candidates before and after the therapy. Abnormal motion is quantified locally in time and space using a statistical distance to normality, and changes induced by the therapy are related with clinical measurements of CRT outcome. Results correlate with recent clinical hypothesis about CRT response, namely that the correction of specific mechanisms responsible for cardiac dyssynchrony conditions the response to the therapy.

Increased resolution in cardiac Magnetic Resonance Imaging (MRI) and growing interest in the effect of small structures in electrophysiology of the heart pose new challenges for cardiac atlases. In this paper we discuss the limitations of current atlas-building models when trying to incorporate cardiac small structure and argue for the need of developing a standard coordinate system for the heart that separates this from the macro-structure common to all individual hearts, in a way analogous to the stereotactic coordinate system from brain atlases. With this goal, we propose a set of methods to obtain two descriptors of the ventricular macro-structure that can be used to build a standardized reference frame: the central curve on the Left Ventricle cavity and the smoothed internal envelope of the Right Ventricle crest (i.e. the curve in the endocardial surface marking the junction between the right ventricular free wall and the septum).

Automatic segmentation of the heart’s left atrium offers great benefits for planning and outcome evaluation of atrial ablation procedures. However, the high anatomical variability of the left atrium presents significant challenges for atlas-guided segmentation. In this paper, we demonstrate an automatic method for left atrium segmentation using weighted voting label fusion and a variant of the demons registration algorithm adapted to handle images with different intensity distributions. We achieve accurate automatic segmentation that is robust to the high anatomical variations in the shape of the left atrium in a clinical dataset of MRA images.

Model-based interpretation of the complex clinical data now available (shape, motion, flow) can provide quantitative information for diagnosis as well as predictions. However such models can be extremely time consuming, which does not always fit with the clinical time constraints. The aim of this work is to propose a model reduction technique to perform faster patient-specific simulations with prior knowledge built from simulations on an average anatomy. Rather than simulating a full fluid problem on individual patients, we create a representative ‘template’ of the artery shape. A full flow simulation is carried out only on this template, and a reduced model is built from the results. Then this reduced model can be transported to the individual geometries, allowing faster computational analysis. Here we propose a preliminary validation of this idea. A well-posed framework based on currents representation of shapes is used to create an unbiased template of the pulmonary artery for 4 patients with Tetralogy of Fallot. Then, a reduced computational fluid dynamics model is built on this template. Finally, we demonstrate that this reduced model can represent a specific patient simulation.

Cardiac resynchronization therapy (CRT) can be an effective procedure for patients with heart failure but 30% of patients do not respond. This may be partially caused by the sub-optimal placement of the left ventricular (LV) lead. Detailed cardiac anatomy and dyssynchrony information could improve optimal LV lead placement. As a pre-interventional imaging modality, cardiac magnetic resonance (MR) imaging has the potential to provide all the relevant information. Whole heart MR image data can be processed to yield detailed anatomical models including the coronary veins. Cine MR data can be used to measure the motion of the LV to determine which regions are late-activating. Finally, late Gadolinium enhancement imaging can be used to detect regions of scarring. This paper presents a complete software solution for the guidance of CRT using pre-procedural MR data combined with live X-ray fluoroscopy. The platform was evaluated using 7 live CRT cases. For each patient, a detailed cardiac model was generated and registered to the X-ray fluoroscopy using multiple views of a catheter looped in the right atrium. There was complete freedom of movement of the X-ray system and respiratory motion compensation was achieved by tracking the diaphragm. The registration was validated using balloon occlusion coronary venograms. The mean 2D target registration error for 7 patients was 1.3 ± 0.68 mm. All patients had a successful left lead implant.

Management and follow-up of chronic aortic dissections continues to be a clinical challenge due to progressive aortic dilatation. To predict dilatation, guidelines suggest follow-up of the aortic diameter. However, dilatation is triggered by haemodynamic parameters (pressure and wall shear stresses (WSS)), and geometry of false (FL) and true lumen (TL). We aimed at a better understanding of TL and FL haemodynamics by performing in-silico (CFD) and in-vitro studies on an idealized dissected aorta and compared this to a typical patient. We observed an increase in diastolic pressure and wall stress in the FL and the presence of diastolic retrograde flow. The inflow jet increased WSS at the proximal FL while a large variability in WSS was induced distally, all being risk factors for wall weakening. In-silico, in-vitro and in-vivo findings were very similar and complementary, showing that their combination can help in a more integrated and extensive assessment of aortic dissections, improving understanding of the haemodynamic conditions and related clinical evolution.

In this paper, we show that canonical (shape-based) geometries can be endowed to cardiac structures using tubular coordinates defined over their medial axis. We give an analytic formulation of these geometries by means of B-Splines. Since B-Splines present vector space structure PCA can be applied to their control points and statistical models relating boundaries and the interior of the anatomical structures can be derived. We demonstrate the applicability in two cardiac structures, the 3D Left Ventricular volume, and the 2D Left-Right ventricle set in 2D Short Axis view.

Left atrium segmentation and the extraction of its geometry remains a challenging problem despite of existing approaches. It is a clinically-relevant important problem with an increasing interest as more research into the mechanism of atrial fibrillation and its recurrence process is undertaken. Contrast-Enhanced (CE) Magnetic Resonance Angiography (MRA) produces excellent images for extracting the atrial geometry. Nevertheless, the variable anatomy of the atrium poses significant challenge for segmentation. To overcome the inherent difficulties with this segmentation, we propose a technique that utilizes the Voronoi subdivision framework for the segmentation. In addition, the segmentation is based on the minimization of a Markov Random Field based energy functional defined within the Voronoi framework. The method also incorporates anatomical priors in the form of a probabilistic atlas. We show how the model is efficient in segmenting atrium images by comparing results from manual segmentations.

We developed an interactive tool for biventricular function analysis from cardiac magnetic resonance (MR) images based on the guide point modelling (GPM) approach [1]. First we built a deformable model of both ventricles of the human heart which consisted of 138 nodes and 82 hexahedral elements, each with bicubic-Bézier-linear interpolation. The model was fitted to a digitized human data set for use as the prior shape in the GPM scheme, which we modified to have a ‘predictor’ step that used a host mesh fitting algorithm [2] to generate predicted points (PPs) based on the user-defined guide points (GPs). Then the model was fitted towards both GPs and PPs through linear least square minimization. The inclusion of the PPs significantly improved the numerical stability of the linear least square fit and significantly accelerated the solution time. This methodology requires further validation for future application in clinical biventricular analysis.

It is well-established that ventricular hypertrophy is a transitional phase through the development of heart failure. A hypertrophic heart can remodel to compensate the loss of pump function, but it eventually becomes incapable of working efficiently, leading to heart failure. Many heart failure patients have preserved pump function (e.g. normal ejection fraction), but increased LV wall thickness and sometimes increased LV mass, which can mask a decrease in contractility. Alterations in the myofibre structure and myocardial material properties can potentially account for the progression of heart failure. We have developed a canine LV finite element model to investigate the effect of ventricular size, myocardial passive material properties, and cardiac contractility on the LV mechanical performance. By comparing mechanical function of normal and abnormal LVs, due to dilation and/or loss of anisotropy and/or reduced contractility, we found that dilation and compromised muscle contractility decreased most indices of cardiac performance. This modelling framework provides insight into the underlying mechanisms of heart failure.

We have recently presented the dynamic deformable elastic template (DET) model for the retrieval of personalised anatomical and functional models of the heart from dynamic cardiac image sequences. The dynamic DET model is a finite element deformable model, for which the minimum of the energy must satisfy a simplified equation of Dynamics. It yielded fairly accurate results during our valuation process on a 45 patients cine MRI database. However, it experienced difficulties when dealing with very large thickening throughout the cardiac cycle, or on highly pathological cases. In this paper, we introduce prescribed displacements as low level constraints to locally drive the model. Non prescribed contour nodes are displaced according to a combination of forces extracted from prescribed points and image gradient. Prescribing a few points in a whole sequence allows us to retrieve much better segmentations on rather difficult cases.

Global and regional cardiac deformation provides important information on myocardial (dys-)function in a variety of clinical settings. Recent developments in the field of echocardiography have allowed the cardiologist to quantify cardiac deformation in a non-invasive manner. Unstitched volumetric data can be captured in a high frame rate by real-time ultrasound imaging. However, most existing methods for measuring myocardial mechanics are often limited to measurements in one or two dimensions. Since myocardial tissue is virtually incompressible, the ventricular wall contains the same volume during the cardiac cycle and, thus, deforms in three dimensions. In this paper, we propose an automatic method to estimate the regional 3D myocardial mechanics on ultrasound images by recovering the 3D non-rigid deformation of the myocardium. The key advantage of our method is fusing multiple information, such as speckle patterns, image gradients, boundary detection, and motion prediction, to achieve a robust tracking on 3D+t ultrasound data. Preliminary results in both in-vitro and in-vivo experiments confirmed these findings in a quantitative manner, as the motion and mechanical parameters, such as displacement and strain, estimated by our method are close to both the ground-truth data and the clinical evaluation. The proposed method is efficient and achieves high speed performance of less than 1 second per frame for volumetric ultrasound data.

Impaired systolic ventricular function is common in patients diagnosed with heart failure (HF) or ischaemic heart disease. The diminished contractile performance with impaired contractility (systolic HF) can be induced by impaired filling function (diastolic HF) and the wall stress (both passive and active) may indicate the progression from diastolic HF to systolic HF. In order to better understand the distribution of active stress during ventricular contraction, a left ventricular (LV) finite element (FE) model incorporating LV fibre geometry and function was developed to parameterise a time-varying model of myocardial contraction by simulating LV mechanics. During systole, the isometric active stress monotonically increased to 95 kPa, and rapidly recovered during isovolumic relaxation. We also observed regional variations of the fibre length dependent contractile stress throughout the LV. The time-varying active stress curve thereby obtained enabled quantification of heart muscle performance. This type of integrative modelling enables the investigation of LV mechanics on an individualised basis.

Because of the intrinsic physiological coupling between the motion and the electrical activity of human heart and available higher resolution imaging sequences, we believe that image-derived cardiac kinematic measurement should be able to reflect patient-specific propagation of cardiac transmembrane potential (TMP). Therefore, in this paper we developed a model-based filter framework, which can recover cardiac electrical activity from MR image sequences. In this particular implementation, the cardiac electro-mechanical coupling process will be properly modelled over a meshfree particle representation of cardiac volume and its fiber structure, and then a model-based unscented Kalman filter (UKF) will be created to incorporate an electro-mechanical coupling model into the state space equation to estimate cardiac electrical activity from MR image sequences. At the end, we not only investigate the performance of our algorithm through two synthetic motion data sets, which are generated by healthy and diseased propagation patterns in an authentical cardiac geometry respectively, but also show the potential usage of our algorithm in clinical diagnosis through a test of one clinical MR image sequence.

Despite advances in both medical image analysis and intracardiac electrophysiological mapping technology, the understanding of cardiac mechano-electrical coupling is still incomplete. This knowledge is of high interest since it would help estimating the cardiac electrophysiology function from the analysis of widely available cardiac images, such as 3D echocardiography. This is important, for example, in the evaluation of the cardiac resynchronization therapy (CRT) where the placement and tuning of the pacemaker leads plays a crucial role in the outcome of the therapy. This paper proposes a method to estimate activation times of myocardium using a cardiac electromechanical model. We use Kernel Ridge Regression to find the relationship between the kinematic descriptors (strain and displacement) and the contraction force caused by the action potential propagation. This regression model is then applied to two 3D echocardiographic sequences from a patient, one in sinus rhythm and the other one with left ventricle pacing, for which strains and displacements have been estimated using incompressible diffeomorphic demons for non-rigid registration.

The use of anti-arrhythmic drugs is common to treat heart rhythm disorders. Computational modeling and simulation are powerful tools that can be used to investigate the effects of specific drugs on cardiac electrophysiology. In this work a patient-specific anatomical heart model is built to study the effects of dofetilide, a drug that affects IKr current in cardiac cells. We study the multi-scale effects of the drug, from cellular to organ level, by simulating electrical propagation on tissue coupled cellular ion kinetics for several heart beats. Different cell populations configurations namely endocardial, midmyocardial and epicardial are used to test the effect of tissue heterogeneity. Results confirmed the expected effects of dofetilide at cellular level, increasing the action potential duration. Pseudo-ECGs obtained for each heart beat correlated well with cellular results showing prolongation of QT segment. These techniques can be applied over the development of more complex drugs that affect multiple cellular currents.

Noninvasive cardiac electrophysiological imaging (IECG), the effort to use body surface potential measurement to estimate subject-specific electrophysiological activity of the heart, traditionally is performed on detailed heart-torso models that are completely reconstructed from a large amount of images. This geometrical modeling brings high demands of operational time and data acquisition, rendering current IECG techniques clinically impractical. In this study, we investigate the feasibility to use an alternative geometrical model that excludes local details but captures subject-specific global geometrical parameters that have been regarded essential for reliable IECG solutions. This is done by using limited images and image metadata to customize a pre-defined, generic ventricle and electrode-array representation to subject-specific ventricle size, position, orientation and electrode position on the body surface. We apply this simplified geometrical modeling in IECG studies of post myocardial infarction patients; the results of transmembrane potential imaging and infarct quantitation are compared with the gold standard and results from the same IECG approach using traditional, detailed heart-torso model. This study shows that local geometrical details do not have significant impact on IECG solutions and excluding them from geometrical modeling might be of potential to drive cardiac electrophysiological imaging closer towards clinical practicability.

We present the fully automated pipeline we have developed to obtain electrophysiological simulations of the heart on a large atlasbased virtual population. This virtual population was generated from a statistical model of left ventricular geometry, represented by a surface model. Correspondence between tetrahedralized volumetric meshes was obtained using Thin Plate Spline warps. Simulations are based on the fast solving of Eikonal equations, and stimulation sites correspond to physiological activation. We report variations of total activation time introduced by geometry, as well as variations in the location of last activation. The obtained results suggest that the total activation time has a strong dependence on LV geometrical variation such as dilation-tohypertrophy.

Model customisation to represent specific experimental or clinical cases is becoming increasingly important as simulations aim to characterise individual variability under physiological and pathological conditions. This study presents a new methodology to customise and regularise heart shape and fibres using imaging data (MRI and DT-MRI). The effect of using generic conductivity tensor values in electrophysiology simulations on these customised meshes is investigated. Simulation results demonstrate the ability of generic parameters to approximate epicardial activation patterns in healthy porcine hearts. Results also show a limited sensitivity of electrical activation times to the anisotropy of these parameters.

Computational imaging of personalized cardiac electrophysiology has attracted increasing research interest because of its clinical relevance in aiding in the diagnosis and prediction of cardiac electrical malfunctions of individual subjects. We have developed a statistical physiological-model-constrained framework that, rather than delivering a personalized cardiac electrophysiological model with customized parameters, uses simple standard electrophysiological models as constraints and produces maximum a posteriori estimation of three-dimensionally distributed transmembrane potential (TMP) dynamics inside the ventricular myocardium of individual subjects [1]. Taking part in 2010 Cardiac Electrophysiological Simulation Challenge (CESC’10), we modify this framework to use epicardial optical mapping data to estimate subject-specific TMP dynamics inside the 3D myocardium. Results of estimated dynamics are compared to the simulations by the same electrophysiological model with standard or adjusted parameters. As shown, while it is rather challenging to personalize the parameters of a cardiac electrophysiological model for the entire 3D myocardium, because of the drastically simplified model structure and limited subject’s data, the presented approach of TMP estimation is able to computationally reproduce subject-specific electrical functions inside the 3D myocardium with simple standard model as constraints.

We present a pipeline for the personalization of model-based Purkinje fast conduction system using fast electrophysiological models and optical mapping data acquired from ex-vivo porcine hearts. The regional density of the Purkinje terminals as well as the latest endocardial activation time were the parameters personalized in an iterative procedure maximizing the similarity between the outcome of the electrophysiological simulations and measurements obtained from optical mapping data. We used a fast wave-front Eikonal-based electrophysiological model that generated the depolarization time maps that were subsequently compared with measurements at each iteration of the optimization stage. The pacing site given by the experimental data and the optimized Purkinje system were introduced into the electrophysiological model. We obtained a regional distribution of Purkinje end-terminals in agreement with findings in the literature. Nevertheless, remaining differences between simulations and measurements after personalization suggest that epicardial data obtained from optical mapping data might not be sufficient to optimize the Purkinje system, which is basically located at the endocardium. On the other hand, the developed pipeline could also be used with endocardial data on electrical activation provided by non-contact or contact mapping system.