P.139 Random Neural Network Features in Patients with Aggressive Multiple Sclerosis Undergoing Autologous Hematopoietic Stem Cell Transplant

Abstract

Background: Objective markers of disease progression are needed for patients with multiple sclerosis (MS). Increased randomness in neural networks is hypothesized to be an important cause of morbidity that can be objectified using graph theory. Methods: We use voxel-based structural similarity determined from T1-weighted MRI scans of 23 patients with MS receiving autologous stem cell transplant (ASCT) to compute cortical covariance network parameters. We examine associations between measures of cortical integration or segregation and biochemical/clinical measures of cortical health or function using Spearman correlation coefficients. P<0.05 was considered significant. Results: Path length increase was associated with markers of greater inflammation (ρ=0.56,P<.046) at baseline and reduced Naa/Cr ratio (P<.041) at 12 months. Reduced lambda was associated with markers of greater grey matter atrophy (ρ=0.55,P<.019) after 12 months and lower cognition (ρ=0.56,P<.008) at 12 months. Reduced clustering was associated with higher neurofilament (ρ=-0.68,P<.010) at baseline, greater white matter atrophy (ρ=0.62,P<.006) after 12 months, lower 2-second PASAT performance (ρ=0.56,P<.011) at baseline, and reduced Naa/Cr ratio (P<.001) at 12 months. Conclusions: Reduced cortical integration and segregation (random network features) co-occur with unfavourable markers of cortical health and function in patients with MS receiving ASCT. Network features show promise as important longitudinal markers of patient status and progression.