Brain Tumor Segmentation and Classification approach for MR Images Based on Convolutional Neural Networks

Abstract

Brain tumor considers one of the most dangerous types of cancers as it affects the main nervous system of the human body. Therefore, several techniques using computer vision has been proposed for early diagnoses and avoiding surgical intervention. However, these techniques face challenges in terms of segmentation and classification processes to detect the brain tumor within Magnetic resonance images (MRI). This paper proposes an automated system for detecting and classifying the brain tumor. This system composed of three phases is including enhancement, segmentation, and classification. The enhancement phase utilizes the Adaptive Histogram Equalization (AHE) in order to adjust the MRI images. The segmentation process was achieved using U-NET to segment the abnormal cells from normal brain tissue. The classification phase was conducted using 3D-CNN to classify the brain tumor into a High-GradeGlioma (HGG) and a Low-Grade-Glioma (LGG). Several experiments were conducted to validate the developed system using Brats-2015 dataset. The system achieved 99.7% (Dice Similarity Coefficient) DSC as an accurate rate for segmentation, and 96% and 98.5% an accurate rate using 5- fold and 10-fold, respectively.