Statistical Textural Feature and Deformable Model Based MR Brain Tumor Segmentation

Segmentation of abnormalities is one of the main focus in medical image processing field for the purpose of diagnosis and treatment planning. The work put forth in this paper has proposed and implemented a semi-automatic technique that yields appropriate segmented regions from MR brain images. The Segmentation technique here utilizes fusion of information beyond human perception from MR images to develop a fused feature map. The information beyond human perception include second order derivatives that are computed from an image which are discussed in detail in relevant section of this paper. This obtained feature map acts as a stopping function for the initialized curve in the framework of an active contour model to obtain a well segmented region of interest. The results obtained from our segmentation method are compared with ground truth segmentation results obtained from experts manually using Jackard's Co-efficient of Similarity and Overlap index. The results obtained on various case studies like Craniophryngioma, High grade Glioma and Microadenoma show a good efficacy of the overall method.