Statistical texture analysis of MRI Images to classify patients affected by Multiple Sclerosis

This paper proposes a vision based system for helping neurologists in the diagnosis of multiple sclerosis (MS), by analyzing textures of the white matter extracted in T2 MRI (Magnetic Resonance Imaging) images. The proposed system consists of three connected subsystems: 1) Pre-processing, 2) Statistical Features Extraction and 3) MRI Slice Classification. The lesions are extracted, in the first step, by using image processing techniques such as morphological and edge detection filters. The diagnosis of the considered MRI slice is based on the analysis of the texture features computed on the white matter highlighted by the first module. More in detail, the statistical moments of the image gray levels are evaluated and are further used by the classification module implemented by using a MLP neural network. The proposed method was tested on a set of 250 simulated MRI images and on a set of 20 real patients achieving very promising results both in terms of sensitivity and in terms of specificity.