Brain lesion segmentation using fuzzy C-means on diffusion-weighted imaging

This paper presents an automatic segmentation of brain lesions from diffusion-weighted imaging (DWI) usingFuzzy C-Means (FCM) algorithm. The lesions are acute stroke, tumour and chronic stroke. Pre-processing is applied to theDWI for intensity normalization, background removal and enhancement. After that, FCM is used for the segmentationprocess. FCM is an iterative process, where the process will stop when the maximum number of iterations is reached or theiteration is repeated until a set point known as the threshold is reached. The FCM provides good segmentation result inhyperintensity and hypointensity lesions according to the high value of the area overlap, and low value of false positive andfalse negative rates. The average dice indices are 0.73(acute stroke), 0.68 (tumour) and 0.82 (chronic stroke)