Non-invasively Grading of Brain Tumor Through Noise Robust Textural and Intensity Based Features

Abstract

Identifying the tumor grade has an important role in surgery planning. A medical practitioner assesses the tumor grade through a biopsy. However, a biopsy usually leads to discomfort and post-biopsy complications. In this work, we propose a noninvasive method to detect the tumor grade using Magnetic Resonance Imaging (MRI). The grading of brain tumor is done through four stages: Skull stripping, brain MRI segmentation, feature extraction, and classification. We have used spatial fuzzy C-means for the segmentation of brain. For textural features extraction, we have used noise robust Local Frequency Descriptor (LFD). The classification is done using random forest, Support Vector Machine (SVM), and decision tree classifiers. The experimental results on a real brain MRI dataset confirm the effectiveness of the proposed method.