Brain MR Image Tumor Segmentation with Ventricular Deformation

Abstract

This paper addresses the issue of the weak association between brain MRI intensity value and anatomical meaning of MR image pixels. By investigating the deformation on brain lateral ventricles and compression from tumor, the correlation between them is quantified and utilized. With the proposed feature extraction component, lateral ventricular deformation is transformed into an additional feature for brain tumor segmentation. Some comparative experiments using both supervised and unsupervised pattern recognition segmentation methods show the improved tumor segmentation accuracy in some image cases.