TitleAutomatic segmentation of the heart from MRI. Is the problem solved ?
ConferenciantAlain Lalande, Fabrice Meriaudeau
Professor/a organitzador/aDomènec Savi Puig Valls
InstitutionUniversity of Bourgogne Franche-Comté, Dijon, France
Date 27-09-2019 12:00
SummaryDelineation of the left ventricular cavity and myocardium (and right ventricle for kinetic study) from cardiac magnetic resonance images (from multi-slice 2-D cine MRI or delayed enhancement MRI (DE-MRI, acquisition 10 minutes after injection of contrast agent)) is a common clinical task to establish diagnosis. The automation of this task has thus been the subject of intense research over the past decades.
Cine-MRI allows the evaluation of the anatomy and the kinetics of the cardiac cavities. Then, the first part of this presentation tackles the automatic segmentation of the cardiac cavities from cine-MRI from the "Automatic Cardiac Diagnosis Challenge" dataset (ACDC), the largest publicly available and fully annotated dataset for the purpose of cardiac MRI (CMRI) assessment. The dataset contains data from 150 multi-equipments CMRI recordings with reference measurements and classification from two medical experts. The overarching objective was to measure how far state-of-the-art deep learning methods can go at assessing CMRI, i.e., segmenting the myocardium and the two ventricles as well as classifying pathologies. In the wake of the 2017 MICCAI-ACDC challenge, results show that the best methods faithfully reproduce the expert analysis, leading to a mean value of 0.97 correlation score for the automatic extraction of clinical indices and an accuracy of 0.96 for automatic diagnosis. These results clearly open the door to highly accurate and fully automatic analysis of cardiac CMRI. But we also identify scenarios for which deep learning methods are still failing.
MRI acquired several minutes after injection of a contrast agent (DE-MRI) is a method of choice to evaluate the extent of fibrosis corresponding to diseased area (for example in case of myocardial infarction), and by extension, to assess viable tissues after injury. The second part of this presentation is dedicated to the post-processing of DE-MRI. The main objective is to automatically detect the different relevant areas (the myocardial contours and the diseased area) and then to make a quantification of the disease, in absolute value (mm3) or percentage of the myocardium. The segmentation and quantification methods would be based on deep learning approaches. Latest architectures using encoder decoder associated with Residual blocks as well as Squeeze and Expandable block will be presented for the myocardium segmentation. The disease areas are then extracted via a set of 2.5 D architectures. At the current time, myocardium segmentation is achieved with an accuracy of 0.98.
Fabrice Mériaudeau (University of Burgundy - France) is a Professor at Université de Bourgogne, Department Electrical Engineering. He received both the master degree in physics at Dijon University, France as well as an Engineering Degree (FIRST) in material sciences in 1994. He obtained a Ph.D. in image processing at the same University in 1997. He was a postdoc from June 1997 to August 1998 at The Oak Ridge National Laboratory and have been collaborating with it since. He was the Director of the Le2i (UMR CNRS), from 2011 to 2016. From 2016 to 2018, he moved to Malaysia and was a professor at Universiti Teknologi PETRONAS (UTP), Dept Electrical and Electronic Engineering. While at UTP, he also held the position of director of the Institute Health and Analytics. His current research interests are on machine learning/AI, image processing for medical/biomedical imaging. He coordinated an Erasmus Mundus Master in Computer Vision and Robotics from 2006 to 2010 and participated to create a new Erasmus Mundus Master in Medical Imaging (maiamaster.udg.edu/) before joining UTP. He was the Vice President for International Affairs for the University of Burgundy from 2010 to 2012. He has authored and co-authored more than 300 international publications and holds four patents.
Since 2000, Alain Lalande is Associate Professor in biophysics and medical image processing in the University of Burgundy, Dijon, France and belongs to the ImViA laboratory (Image and Artificial Intelligence). He received the Ph.D. degree in biophysics and medical image processing from the University of Burgundy, Dijon, France, in 1999. His research interests concern the medical imaging domain, more specifically the imaging of the cardiovascular system (in particular from MRI), and cover the topics of image acquisition, image post-processing, and protocol design. He is the head in the development of the QIR software (dedicated of the post-processing of the cardiovascular MRI and that is currently follow a process of technology transfer with the CASIS company). He has authored and co-authored of around 70 international publications in journal with impact factor and get a h-index of 24 according to Google Scholar.