Probabilistic determination of brain tumor locations

Abstract

Determination of the origin of tumor locations is an important issue in terms of diagnosis and treatment of patients with glioma. The aim of this study was to determine and analyze a probabilistic brain tumor map reflecting tumor observation frequencies in different brain locations. T1-weighted MR images of 232 patients diagnosed with high and low grade brain tumors were analyzed. The data were collected from both online MRI brain tumor data resources shared for academic usage as well as brain tumor patients from Baskent University Hospital. We obtained 78 high and 54 low-grades MRI scans from Baskent University Hospital, and 20 high and 10 low-grades images from the MICCAI 2012 Challenge on Multimodal Brain Tumor Segmentation (BRATS). Combining all these MRI scans together with the use of brain imaging techniques created the probabilistic brain tumor map. Following brain extraction, image registration is implemented to transfer all MRI scans to the reference image coordinates, and individual transformation matrices are obtained for each data. Then brain tumors are segmented, and a radiologist confirmed the segmentation results. Then segmented images are registered to the standard coordinates by multiplying them with their specific transformation matrices. Our results indicated that there is a difference between high and low-grade tumor regions. Many of tumor positions were around the frontal and temporal ventricular zone for high-grades while the low-grade tumors were located around the posterior ventricular wall. These findings also support the theory that there is a close relationship between gliomas and ventricle region where neural stem cells are emanated.|Keywords : MR, Brain Image Analysis, Registration, Segmentation, Brain Tumor, Glioma, Neural Stem Cell, Probabilistic Map.