Fusion of multiparametric and longitudinal MRI data by representation learning

Full description and practical details in the attached PDF.

We target the characterization of patient groups and potential patterns in patients with chronic non-ischemic cardiomyopathies, using representation learning. Our specific aim is to develop models that can incorporate multi-scale and heterogeneous data, as well as the dynamic nature of the data over time. To address this, we will deploy data-driven (statistical) models based on various biomarkers extracted from imaging. These will be in the longer term fused with virtual electrical & mechanical models (developed in other work packages of the “ChroniCardio” project) to predict the risk of sudden cardiac death, arrhythmia, and heart failure. In this PhD, we will target the fusion of multi-scale / multi-modal data both from scalar variables extracted from high-dimensional MRI data, and high-dimensional variables (e.g. pixelized maps of myocardial patterns). Specifically, we will: align the data to a common reference by adapting computational anatomy tools available at CREATIS / develop multi-view and fusion learning models for the statistical analysis / model the dynamics of multi-scale features from longitudinal studies.

Location: Institut de Mathématiques de Marseille, joint PhD with CREATIS Lyon.

Duration: 3 years, starting September-October 2023