Federated causal inference to link structural brain damage to cognitive impairment in multiple sclerosis

An important and long-standing problem in multiple sclerosis (MS) is the clinico-radiological paradox:
the lack of a strong link between brain damage as seen on brain images and the physical and
cognitive status of a patient. Cognitive impairment affects one out of two patients and has important
repercussions on daily life activities.
In this project, we will use artificial intelligence (AI) to find imaging biomarkers for the cognitive
state of people with MS. The first use of AI is to perform a smart brain image compression,
synthesising the complex brain image information in a reduced set of variables. Then, we will train
an AI model to predict the cognitive state of MS patients from this compressed image, using a
technique called federated learning, allowing to train the model without sharing data between clinical
centres. Lastly, we will investigate whether causal inference modelling, a third AI technique, can
improve the prediction in the second step, and uncover hidden causal relationships to shed new light
on the paradox.
Besides methodological advances in combinations of AI techniques, the clinical advances of this
project are [1] new, causal insights in the paradox and [2] a model that pre-selects patients in need
of rigorous neuropsychological testing from routine brain imaging during clinical follow-up, guiding
clinical staff to earlier intervention to prevent further deterioration.