Automated segmentation of brain lesion based on diffusion-weighted MRI using a split and merge approach

Abstract

This paper presents a segmentation of brain lesion from diffusion-weighted magnetic resonance images (DW-MRI or DWI) using a split and merge approach. The lesions are hyperintense lesion from tumour, acute infarction, haemorrhage and abscess, and hypointense lesion from chronic infarction and haemorrhage. Pre-processing is applied to the DWI for intensity normalization, background removal and intensity enhancement. Then, the split and merge algorithm is designed to segment the lesion. Histogram thresholding technique is used at each split level to detect pixels with either hyperintense or hypointense. Several statistical features are discussed and evaluated to select the best feature as homogeneity criteria. The analysis shows that mean and number of lesion pixels are the best homogeneity criteria. Hyperintense and hypointense lesions are segmented automatically by merging the regions that are homogenous according to the criteria.