GAMER MRI : a deep dive in multiple sclerosis pathology and clinical disability

Multiple sclerosis (MS) is a chronic inflammatory disease of the central nervous system. Typical characteristics are multifocal inflammatory infiltration, demyelination, remyelination, and axonal loss in the microenvironment of brain tissue. The advanced MRI (aMRI) sequences and the quantitative measures derived from them can provide surrogate measurements on these microstructural changes. The information provided by aMRI is abundant, but partially redundant among them. There is a need to assess which aMRI or quantitative measures are important to a given task and explore the benefit of considering them jointly in studying MS axonal/myelin damage and repair.
We proposed and validated a novel method based on the convolutional neural network and gated attention mechanism (GAMER-MRI) in the application of well-understood stroke-related and multiple sclerosis-related lesion classification. The method gave an attention weight-based importance order of MR contrasts in line with clinical understanding. Next, we extended the method to tackle highly intercorrelated diffusion measures based on diffusion MRI in the classification of MS lesion and perilesional tissue. The correlation of selected measures with patient-level measures including the clinical scale of movement disability and the biological measure on the degraded axons were statistically significant, and the combinations of them had a stronger correlation. Last, we demonstrated the improvement of the method on the patient-level classification and a proposed approach to identify the brain regions contributing towards the importance of the images through the combination of the relevance maps and the corresponding attention weights.
Along with these developments, we demonstrated that GAMER-MRI was able to give us the importance of MR images from the local lesion-level analysis to the global patient-level analysis and be a new means to jointly combine the abundant information in different kinds of MRI images for a more comprehensive analysis in the future.