Project Details
Projekt
Graph Learning and Pathophysiological Models: Towards a New Classification of White Matter Lesions in Multiple Sclerosis
Applicants
Professor Dr. Mark Mühlau; Professor Dr. Daniel Rückert, Ph.D.; Privatdozent Dr. Benedikt Wiestler
Subject Area
Nuclear Medicine, Radiotherapy, Radiobiology
Term
since 2019
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 428223038
Multiple Sclerosis (MS) is the most common chronic inflammatory disease of the central nervous system and affects over 2.5 million people worldwide. To date, MS is the leading non-traumatic cause of serious neurologic disability in young adults. MR imaging is a mainstay in the clinical management of MS patients: By detecting subclinical disease activity, MR opens an opportunity to adapt therapy before overt clinical deterioration. However, in stark contrast to the advances in understanding the complex pathophysiology underlying MS and development of targeted therapies, MR imaging has remained focused on simple measures of disease activity, i.e. the occurrence of new T2-hyperintense white matter lesions. However, several pathological studies convincingly demonstrated that white matter lesions are by no means mere scars resulting from faded acute inflammatory demyelination, but reflect different pathophysiological processes, and therefore hold valuable information. Unfortunately, this wealth of information contained in MR imaging data to date has not been used in clinical practice. To uncover these disease- and patient-specific information contained in MR images, we will build upon and expand our achievements during the first funding period of the DFG Priority Programme “Radiomics”. During the first funding period, we established a powerful pipeline for cross-sectional lesion segmentation and feature description. Continuing this work, we aim to synergistically identify data- and hypothesis-driven biomarkers of MS in MR images with a special focus on longitudinal changes to better capture pathophysiological processes. We will therefore in parallel work on the development of (self-)supervised lesion descriptions and hypothesis-driven models of MS lesion classes. Importantly, there will be a close feedback loop between both, as we also aim to validate findings from one with results from the other. For example, we will investigate if lesions belonging to a particular pathophysiological group also share common imaging characteristics. Relying on our prior work on Graph Convolutional Networks, we will combine information from these two approaches into patient-level graphs of MS lesions, enabling us to model the imaging data in unprecedented ways and power a detailed investigation and analysis of longitudinal information. We will rely on our large, well characterized in-house prospective observational cohort of MS patients (> 3000 patients) for discovery and additionally validate our findings in cooperation with other centers (Mainz, Zürich). In this project, we will identify MRI biomarkers of MS pathophysiology that can be extracted from routine scans and made available to clinical practice relatively easily. This project will thereby highlight the potential of radiomics-based image analysis and contribute to supporting clinical decision making in MS.
DFG Programme
Priority Programmes
International Connection
Switzerland
Cooperation Partner
Professor Dr. Björn Menze
