MS lesions segmentation using K-medoids and statistical features

Lesions inside white matter of brain are the pathological symbols of MS. These lesions are noticed well in MRI images. Determining and quantitative analysis of these lesions has been an important yet challenging task in recent year. In this paper, we propose a new method for segmenting these lesion in noisy T2-weighted MRI images based on statistical texture features and a variation of conventional k-means clustering, namely, k-medoids clustering algorithm. We compared the performance of our method with two well-known methods in terms of dice metric, specificity and false negative rate (FNR). Results proved outperformance our method in segmenting noisy MRI images.

Keywords:

Segmentation , K-medoids , Dice metric , Specificity , False negative rate