BRAIN TUMOR SEGMENTATION AND CLASSIFICATION USING CNN PRE-TRAINED VGG-16 MODEL IN MRI IMAGES

Abstract

The formation of a group of abnormal cells in the brain that penetrate the neighboring tissues is known as a brain tumor. The initial detection of brain tumors is necessary to aid doctors in treating cancer patients to increase the survival rate. Various deep learning models are discovered and developed for efficient brain tumor detection and classification. In this research, a transfer learning-based approach is proposed to resolve overfitting issues in classification. The BraTS – 2018 dataset is utilized in this research for segmentation and classification. Batch normalization is utilized in this experiment for data pre-processing and fed to a convolutional layer of CNN for extracting features from Magnetic Resonance Images (MRI). Then, an Adaptive Whale Optimization (AWO) algorithm is utilized to select effective features. This work proposes a Convolutional Neural Network (CNN) based segmentation and a transfer learning-based VGG-16 model for effective classification. The performance of the proposed CNN-VGG-16 technique is analyzed through various tumor regions like TC, ET, and WT. The proposed method attains a Dice score accuracy of 99.6%, 95.35%, and 94%, respectively, when compared to other existing algorithms like CNN, VGG-net, and ResNet.