235 — Segmentation of Multiple Sclerosis lesion in brain MR images using Fuzzy C-Means

Fuzzy c-means (FCM) is a clustering algorithm that allows individual datapoints to belong to more than just one luster. This is very useful when answers must be approximate, or are representative of a probablistic cluster membership.

The authors use this clustering technique to segment MRI brain scans to detect lesions that are signs of white-matter degredation in multiple sclerosis. I’ve read about this problem before in this paper (and there are more MS papers here). This is a challenge, and the current state-of-the-art lesion segmentation approaches all use deep-learning.

This approach instead uses edge detection to determine the boundaries of lesions, and then validates each regions for lesioniness.

The authors found that of the edge-detection methods that they used (Canny, Sobel, and Marr-Hildreth), canny edge-detection worked the best on the low-contrast brain-scans, and performed the best to reduce noise in the segmentation as well.

This preprint doesn’t include too much information on the results, but it will be interesting to see how non-learning methods compete with GPU-heavy compute in a clinical setting.