We investigate the potential advantages of using distribution-valued variables to capture image information instead of the conventional techniques which use mean values to represent entire regions-of-interest. Using distribution-based distances in comparison with a mean-based approach, we explore various models to predict motor performance recovery in stroke patients. Our experiments indicate that for predictor variables Orientation Dispersion index and the Restricted Diffusion Index (diffusion scalars derived from the NODDI model), the distribution representations can lead to significantly improved regression models, over the respective mean values. Additionally, using the baseline Fugl-Meyer scores to predict follow-up Fugl-Meyer scores further enhances the model statistics.
This abstract and the presentation materials are available to members only; a login is required.