Segmentation of sparse annotated data - application to cardiac imaging

Abstract:

In cardiac MR, segmenting the left ventricle, the right ventricle and the myocardium is a common task in clinical routine. Several state-of-the-art deep learning algorithms are able to achieve reliable and great performances for this task. Nevertheless, it is often performed in a supervised way, i.e. annotated data are needed. Because these annotations are time-consuming for the clinician to make, recent works focus on being able to limit the needs of annotation and still provide robust and reliable segmentation. Different strategies exist to overcome this limitation, such as transfer learning or self-supervised learning, which are learning prior knowledge on a similar annotated dataset or without any annotation.

The objective of this project is to be able to provide robust and reliable segmentation of a sparse annotated cardiac MR dataset. The prospective student will develop a segmentation network based on recent strategies for sparsed annotated data and compare them to state-of-the-art deep-learning segmentation methods.

Maxime Di Folco
Maxime Di Folco
Research Scientist

My research interest is the study of the cardiac function via machine learning methods, in particular representation learning methods that aim to acquire low dimensional representation of high dimensional data. I have a strong interest in cardiac remodelling (adaptation of the heart to its environment or a disease), notably the study of the deformation and shape aspects.