Towards the integration of advanced magnetic resonance imaging biomarkers in multiple sclerosis clinical practice

Details

Supervisors

Macq, Benoît ; Maggi, Pietro

Degree label

Master [120] : ingénieur civil en science des données, à finalité spécialisée

Abstract

Multiple sclerosis (MS) is a chronic immune-mediated disease characterized by focal inflammatory lesions in the brain and spinal cord. Magnetic Resonance Imaging (MRI) is an essential tool in the diagnosis of MS. However, markers currently used in the clinic are not specific enough for MS. Several advanced MRI biomarkers, including paramagnetic rim lesions (PRL), more specific to MS and based on advanced MRI techniques have recently been studied. This thesis aims to i) develop a multi-modal database including clinical, biological and imaging features of MS patients; ii) explore the heterogeneity of the disease using statistics and machine learning techniques; iii) to contribute to the collective effort of researchers in the field to automate the extraction of MRI features and iv) to introduce the issue of longitudinal detection of advanced MRI features in MS. The first part of this thesis focuses on the completion of the multi-modal database. It first presents the tools developed to automate the MRI features' extraction: a pipeline for converting from the DICOM format to the BIDS standard as well as a pipeline for the extraction of volumetric data and the localization of lesions. Then, this work shows that RimNet, a 3D U-Net inspired Convolutional Neural Network for automated PRL detection, has a good generalization capacity and that it is able to make longitudinal consistent predictions. This part ends with the specification of a database management system specific to the context along with a proposal of solution. The second part presents the results of statistical analysis, and describes in detail how our database resembles or not the current literature in the field. It also announces the steps still in progress to perfect the objectives, i.e. the techniques of imputation of data to be explored, the regression for the prognosis and the clustering for the detection of the potential subtypes of MS. The impact of this thesis is manifold. First, it has created an innovative database, including clinical, biological and both established and advanced imaging features. On the way to the completion of this database, several tools have been developed which have also made it possible to speed up the work of researchers in this field by automating specific time-consuming tasks. The thesis also lays the foundations for the development of a multi-modal database management system in the medical imaging sector. Finally, the preliminary analysis of this multi-modal database, already helped to get more insights on the heterogeneus nature of the disease in different MS patients.