An efficient scheme for detecting region of interest encompassing the brain tumor from 3D MRI data based on voxel statistics

Abstract

Segmentation of a region containing the brain tumor from 3D magnetic resonance imaging (MRI) data can help physicians to diagnose accurately the size and malignancy of the tumor. However, manual segmentation is time consuming and involves risk of having inaccurate result. In this paper, an automatic method of segmenting the region of interest (ROI), a region encompassing the brain tumor and its neighborhood, is proposed based on voxel statistics. In the proposed method, first possible candidate selection is performed utilizing intensity characteristics of tumor region in the FLAIR and T1 images of MRI data. Next, a cubic shaped 3D mean filtering operation is applied on the whole volumetric data to obtain filtered volume where some random intensity behavior is expected to be eliminated. Finally, from the resulting 3D FLAIR data, ROI is extracted based on cumulative distribution function of intensity. It is found that the extracted ROI offers significant reduction of the overall MRI volume without losing tumor data. The proposed ROI extraction scheme is tested on 20 real life high grade tumor cases obtained from a widely used database and a very satisfactory performance is obtained in terms of segmentation accuracy, overall volume reduction and computational time.