Unlocking the prognostic potential of quantitative magnetic resonance imaging for multiple sclerosis through a multimodal approach

Abstract

Magnetic resonance imaging (MRI) is the most widely used paraclinical tool for diagnosing and monitoring multiple sclerosis (MS). However, given known heterogeneity in pathophysiology, even with robust conventional MRI (cMRI) measures such as white matter lesion (WML) number and volume, the relationship with MS progression is only modest, hindering its prognostic use in MS. One major burden is the lack of characterization of the white matter (WM) space. With recent advances in quantitative MRI (qMRI) techniques, the enhanced sensitivity and specificity of individual qMRI have shed light on its clinical potential. But none of them has yet been validated as a stand-alone biomarker for MS, and their clinical value has not been clearly demonstrated. Therefore, this thesis investigates the potential of a multi-modal qMRI approach for WM characterization and prognostic applications. The thesis work is split into four different phases. The first phase has explored the multimodal relationship between qMRI and volumetric measures across three brain regions relevant to MS – WML, normal appearing white matter (NAWM) and deep gray matter (DGM). The results have demonstrated the clinical relevance of qMRI and the feasibility of the multimodality approach. The second and third phases were then motivated to provide a deeper characterization of the WM by defining finer regions related to WML and perilesional NAWM, both cross-sectionally and longitudinally. These two phases have demonstrated that qMRI metrics not only are sensitive to WM heterogeneity, but also provide information about future WM transitions. The final phase tasked a neural network model in predicting WML evolution using separated and combined inputs from qMRI and cMRI, and assessed the performance. The positive result strongly suggests the potential of qMRI and cMRI to predict WML evolution. These findings indicate richer levels of WM heterogeneity than the binary distinction between WML and NAWM, which can be sensitized using qMRI. While a multimodal qMRI approach helps characterize WM, cMRI also demonstrates the capability to predict WML evolution. Overall, this work sheds light on the clinical value of qMRI in characterizing the WM space, demonstrates the extended utility of cMRI, and lays a solid foundation for future studies in WML prediction, ultimately helping improve MS prognosis.