Segmentation for Multiple Sclerosis Lesions Based on 3D Volume Enhancement and 3D Alpha Matting

Abstract

Segmenting of Multiple Sclerosis (MS) lesions in magnetic resonance (MR) images is a hot issue in biomedical engineering. This paper presents a novel approach for segmentation of MS lesions in T1-weighted (T1-w), T2-weighted (T2-w), and fluid-attenuated inversion recovery (Flair) MR images. The proposed approach is based on three-dimensional (3D) enhancement followed by false positive reduction methods and a three dimensional (3D) alpha matting technique. Firstly, the MS lesions in 3D volumes are enhanced driven by segmenting and enhancing single slices with MS lesions. Then a binary volume of interests (VOIs) of potential MS lesions is generated by thresholding. Secondly, multimodality information is used to segment the brain white matter. Then the location and the size of MS lesions are used to remove false positive VOIs. Finally, a 3D alpha matting method is utilized to refine the segmentation results, and to compute the VOIs with sub-pixel precision by considering partial volume effects. The experiments on real MRI data shows the unsupervised segmentation method can obtain better result than some state-of-the-art methods.