Description
Date depot: 13 janvier 2022
Titre: Machine learning for multimodal neuroimaging in multiple sclerosis
Directeur de thèse:
Olivier COLLIOT (ICM)
Encadrant :
Bruno STANKOFF (APHP (UP))
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Sciences de l’information et sciences du vivant
Resumé: Multiple sclerosis (MS) is a chronic disease of the central nervous system characterized by multi-focal inflammatory and demyelinating lesions disseminated in the brain and in the spinal cord. Conventional magnetic resonance imaging (MRI) techniques are highly sensitive for the detection and quantification of MS plaques in the white matter. However, they are not specific for the mechanisms that may underlie this neurodegeneration, and a large body of evidence has clearly demonstrated that the quantification of lesion number or lesion load on MRI poorly predict subsequent disability worsening, leading to the concept of a clinico-radiological paradox.
The aim of this project is to design and validate deep learning methods for the automatic processing of MRI data in order to further our comprehension of MS mechanisms and patient stratification and to obtain reliable biomarkers.
The first objective will be to develop a deep learning method for the segmentation of the CPs from MRI data and apply it to study large cohorts of subjects. The project will be built upon previous work perform in the ARAMIS lab. The algorithm will then be applied to study large cohorts of subjects. More specifically, we will aim to apply it to larger cohorts of patients with MS in order to shed light on the role of CP in disease evolution. We also propose to apply it to over 50,000 healthy subjects to perform a genome wide association study (GWAS).
The second objective will be to characterize on MRI data, perfusion changes described on PET data and develop deep learning-based methods to explore their effect on the disease. To that aim, we propose to design new generative models based on extensions of generative adversarial networks in order to reveal PET-specific information from MRI data only. Furthermore, we propose to develop methods that can predict the clinical course of patients based on the combination of these different informations.
The project is performed in co-supervision between O. Colliot (DR CNRS) and Bruno Stankoff (PU-PH, neurologist), bringing complementary expertise in machine learning, neuroimaging, clinical and pathophysiological aspects of MS.
Doctorant.e: Yazdan Panah Arya