Detection and Localization of High-Grade Glioma Brain Tumor using Fusion of MRI and MRSI

A brain tumor is majorly classified in two classes: low-grade glioma and high-grade glioma. The existing detection and localization methods of both types of brain tumor use intraoperative or preoperative imaging. The key limitation of intraoperative imaging is the requirement for expensive scanning machine. In addition, patients are exposed to harmful dye content in this type of imaging. To overcome the limitations of intraoperative imaging, preoperative image modalities such as CT scan, MRI, MRSI, PET, and US have been used to locate the brain tumor. However, a single preoperative image modality is not self-sufficient to detect and localize both types of tumors accurately. Furthermore, among the two types, localization of high-grade glioma is of utmost importance for neurosurgeons as it overgrows. On the contrary, detecting and localizing high-grade glioma is challenging because it is diffused in nature. This paper introduces a new method for detection and localization of high-grade glioma brain tumor using a fusion of preoperative MRI and MRSI images. MRI gives soft tissue discrimination, while MRSI preserves biochemical information. Thus, the fusion of these two modalities leads to accurate tumor detection. The proposed method's performance is evaluated employing four standard classifiers on a real-time, self-collected dataset. The experimental results indicate that the random forest classifier achieves 83% accuracy and outperforms other classifiers. After detecting the tumor, the proposed method uses a texton map for localization of the tumorous tissues. All the diffused tumor tissues, from the smallest to the biggest size, are localized.