Amy Kuceyeski
MS is a complicated disease that has a heterogeneous presentation and progression. We believe that quantitative methods, including machine learning, may allow researchers to shed light on the relationship among white matter lesions, gray matter atrophy, and cognitive/physical impairments over the course of the disease. In our paper, we chose to use our Network Modification (NeMo) Tool to investigate the effect of lesions on the structural connectome and connected gray matter regions. This tool allows an estimation of the “disconnectome,” or disruption of the connectome, and does not require diffusion imaging or tractography to be performed in patients who may have additional disease-related noise in the diffusion MRI signal. We found that atrophy in deep gray matter regions was significantly related to the number of lesions in connecting white matter, a result that was replicated by Steenwijk et al1 using similar analytic techniques. In the original paper, we also reported the central role of visual processing areas in processing speed as measured with the Symbol Digit Modalities Test (SDMT). Interestingly, we found the same brain regions to be significantly associated with SDMT scores in subjects with stroke, a result that confirms these regions’ central role in processing speed despite the vast differences in the demographics of the population and the etiology of white matter disruption. We are currently working with several groups to apply the NeMo Tool in clinical, retrospective patient populations (MS and stroke) that either have too much noise in the diffusion signal or do not have diffusion imaging in order to assess the effect of pathologic abnormalities on the structural connectome. Another group used the tool to relate white matter lesions’ impact on the connectome to symptoms of fatigue in patients with MS.2 We have expanded on our AJNR paper by investigating the role of white matter lesions’ disconnectome in predicting future cognitive status in patients with MS using machine learning approaches. We presented preliminary results at a New York Academy of Sciences meeting and currently have a full manuscript under review. We chose to pursue this particular line of research in order to bolster the clinical applicability of our work. Quantitative modeling, in particular machine learning, applied to neuroimaging metrics will be essential in understanding MS, improving the accuracy of prognoses, and possibly even developing effective treatments.