Camara O., et al. (Eds.) Statistical Atlases and Computational Models of the Heart. Imaging and Modelling Challenges

Springer-Verlag Berlin Heidelberg, 2013. — 399 p. — ISBN 978-3-642-36960-5, e-ISBN 978-3-642-36961-2.Third InternationalWorkshop, STACOM 2012, Held in Conjunction with MICCAI 2012, Nice, France, October 5, 2012, Revised Selected Papers.This year’s edition of STACOM 2012 was held in conjunction with the MICCAI conference (Nice, France), and followed the last two editions: STACOM 2010 (held in 2010, Beijing, China) and STACOM 2011 (Toronto, Canada). The main purpose was to provide a forum for the discussion of the latest developments in the areas of statistical atlases and computational imaging and modeling of the heart. This broad aim included: cardiac mapping, image processing, atlas construction, statistical modeling of cardiac function across different patient populations, cardiac computational physiology, model personalization, ontological schemata for data and results, atlas-based functional analysis, integrated functional and structural analyses, as well as the clinical applicability of these methods. STACOM 2012 again drew in participants from around the world, with 42 papers accepted and published by Springer in this volume of Lecture Notes in Computer Science. The integration of cardiac models in preclinical and clinical platforms is important for understanding disease, evaluating the treatment, and planning an intervention. However, significant clinical translation of these tools is constrained by the lack of complete and rigorous technical and clinical validation as well as benchmarking of the developed tools. To validate the models, available groundtruth data capturing generic knowledge on healthy and pathological hearts are required. Several efforts are now established to provide Web-accessible structural and functional experimental datasets for clinical, research, and educational purposes. We believe that these approaches will only be effectively developed through collaboration across the full research scope of the cardiac imaging and modelling communities. In addition to regular papers focused on the main themes, STACOM 2012 was also enhanced by four different challenges for participants to test their computational tools on given data: a Computational Fluid Dynamics (CFD) Challenge (organized by Siemens Corporation, Corporate Technology, King’s College London, and Open Source Medical Software Corporation US), a Delayed Enhancement Magnetic Resonance Imaging (DE-MRI) Segmentation Challenge (organized by King’s College London and KU Leuven), a Cardiac Motion Analysis Challenge (organized by Philips Research Medisys, INRIA, CHU Caena and Cyceron, KU Leuven) and a Cardiac Landmark Challenge (organized by the University of Auckland). Each challenge is briefly described below.
CFD Challenge — Advances in medical imaging and computational fluid dynamics (CFD) techniques make it possible to simulate blood flow and pressure in cardiovascular models built from patient data. In particular, image-based CFD techniques enable the calculation of pressure gradients through an aortic coarctation noninvasively and for various physiological states. However, the wide range of available software, modeling formulations and approaches to boundary condition specification make it difficult to assess the consistency and repeatability of computational predictions. As a first step to ward gauging the ability of CFD methods to make reliable predictions in the clinic, the STACOM 2012 CFD Challenge set out to assess the variability in the calculation of the pressure gradient through a moderate thoracic aortic coarctation model under resting conditions. A geometrical model of the ascending aorta, aortic arch, descending aorta, and upper branch vessels of a young patient was provided to participants. Additionally, inflow boundary condition and flow splits in the branches were also provided, along with systolic and diastolic proximal pressures. The participants were asked to reproduce all the given data and to compute the average and peak pressure gradient through the coarctation. The computed pressure gradient will be compared with a clinically measured surrogate of the pressure gradient through the coarctation: i.e., the difference between pressure cuff measurements taken in the arm (representative of pressure proximal to the coarctation) and in the leg (representative of pressure distal to the coarctation). The challenge attracted six participant groups and these proceedings contain the details of the methodologies and results obtained by each of these participating groups, as well as reflections on the challenges derived from the lack of absolute ground truth data in clinical measurements.
DE-MRI Segmentation Challenge — Delayed-enhancement magnetic resonance imaging (DE-MRI) is a powerful tool for detecting fibrosis/scar in the myocardium of the heart. Recently there has been much interest in quantifying left ventricular fibrosis/scar with many useful applications including cardiac resynchronization therapy guidance, ventricular tachycardia ablation therapy guidance, and investigation of remodeling following myocardial infarction. The STACOM 2012 DE-MRI Segmentation Challenge made available 30 DEMRI datasets to participants for segmentation of enhanced regions from postmyocardial infarction patients (15) and pigs (15) that had been subjected to myocardial ischemia. Of each of these 15 datasets, 5 were provided as training samples with manual annotation of enhanced regions made by experienced observers. The provided data included the DE-MRI ventricle scan and a segmentation of the myocardium. Ground truth was established by using manual segmentations from experienced clinical observers. The aim of the challenge was to create a standard dataset for benchmarking of DE-MRI segmentation algorithms and to compare different state-of-the-art algorithms. The challenge attracted five participants and these proceedings contain the details of the methodologies and results from each of these participating groups. A collation study of this segmentation challenge is also included in these proceedings.
LV Landmark Detection Challenge —Anatomical landmark annotation of the heart plays a significant role in cardiac MR analysis. It enables more robust and accurate functional and structural analysis of the heart. Yet, this prerequisite step for an automated cardiac MRI segmentation method is usually determined by tedious and subjective manual positioning. At the STACOM 2012 workshop, the Cardiac Atlas Project provided 100 cardiac MR images for a fully automated detection of anatomical landmarks of the heart, which include mitral valve, RV insert and LV center points. To help participants develop and evaluat their methods, another 100 patient images were made available for training, accompanied with landmarks determined by expert observers. The training and validation cases were randomly selected from the same cohort. The challenge attracted nine participants of which two submitted their detection results. Details of their methods are presented in these proceedings.
Motion Analysis Challenge —This year, the second cardiac Motion Analysis Challenge (cMAC2) focused on 3D ultrasound. For any modality, validation of myocardial motion and deformation quantification algorithms is a challenging task that requires reference databases with controlled ground truth. In the case of ultrasound, ground truth motion cannot be obtained through manual landmarking of the data. For this reason, the focus was on phantom data, including synthetic images and in vitro acquisitions. The synthetic images proposed in this challenge combined an electromechanical model with an ultrasound imaging model. We provided ten sequences spanning different values of the global conductivity, global contractility, and electrical delay parameters of the 3D electromechanical model. In the current dataset, a single probe design was considered. Scatterers were randomly placed in the myocardial geometry and moved along the cardiac cycle according to the result of the mechanical simulation. The approach accelerated the convolution of a 3D point spread function (PSF) by multiple 1D convolutions and allowed the generation of more realistic ultrasound datasets. These data included ground truth on the motion field so that each challenger can report quantitative results. Ground truth motion was provided as a series of volumetric meshes. For the physical phantom, a Polyvinyl alcohol was produced by repeated freezing and thawing of PVA solution (with addition of graphite) in a cylindrical shape. One extremity of the cylinder was fixed. The other extremity was moved by a piston imposing both rotation and longitudinal compression. Ground truth data for the phantom set-up were captured using microsonometry (which gave the distance between crystals pairs embedded in the gel and is therefore considered as ground truth for radial, circumferential, and longitudinal strains). The cMAC2 challenge attracted 7 participant groups.
We hope that the results obtained by these four challenges, together with all regular paper contributions, will act to accelerate progress in the important areas of heart function and structure analysis.