This work proposes to use scalar features calculated from diffusion MR data alongside structural MR intensities in the automated segmentation of Multiple Sclerosis (MS) lesions. We acquired and processed multi-contrast MR data from 7 MS patients, used random forests to segment lesions, and evaluated our method via DICE scores, achieving scores over 0.65. Finally, we made use of the random forest framework to assess the discriminative power of the estimated features. We show that diffusion features estimated from the diffusion tensor are as discriminative as T1 and T2 intensities for the classification task.     «

This work proposes to use scalar features calculated from diffusion MR data alongside structural MR intensities in the automated segmentation of Multiple Sclerosis (MS) lesions. We acquired and processed multi-contrast MR data from 7 MS patients, used random forests to segment lesions, and evaluated our method via DICE scores, achieving scores over 0.65. Finally, we made use of the random forest framework to assess the discriminative power of the estimated features. We show that diffusion featur...     »