Automatic Segmentation of Diffusely Abnormal White Matter in Multiple Sclerosis using Fuzzy C-means Clustering

Motivation: Diffusely abnormal white matter (DAWM) occurs in ~25% of multiple sclerosis patients. Currently there is no consensus on how to best automatically segment DAWM which is difficult to identify and quantify. Goal(s): Use fuzzy c-means clustering to segment DAWM. Approach: A data set from DAWM+ patients (n=40) was used to train a clustering algorithm on proton-density (PDw) and T2-weighted (T2w) images. This algorithm was tested on 9 patients visually identified as DAWM+ and 51 patients visually identified as DAWM-. Results: Clustering segmented PDw/T2w hyperintensities in DAWM+ test subjects. Patients visually identified as DAWM- also had regions segmented as DAWM. Impact: Unsupervised clustering could provide an unbiased approach to automatically segment diffusely abnormal white matter (DAWM) in multiple sclerosis patients in vivo. In this study, the subtle hyperintensities associated with DAWM were segmented on proton density and T2-weighted images.