Enhancement of brain tumor classification from MRI images using multi-path convolutional neural network with SVM classifier

Abstract

The use of Convolutional Neural Networks (CNNs) has brought significant progress to the field of medical image segmentation and classification. This study aims to explore the potential of CNNs in classifying brain MRI images, which is crucial for faster identification of brain tumor types to speed up the treatment process. Many researchers are working hard to synthesize neural networks with less complexity and higher accuracy. In this regard, we start with a single-path CNN structure from the literature as a basis and halve the number of filters of three convolutional layers down to 4, 8 and 16. Then, we improve the accuracy of brain tumor classification by modifying a multi-path CNN structure constructed from our simplified single-path CNNs. The greatest improvement in the network's accuracy and sensitivity occurred when we replaced the conventional SoftMax classifier with a Support Vector Machine (SVM) classifier. Comparison of the proposed structure to the work of other researchers demonstrates a notable increment in accuracy of brain tumor classification by more than 10 % of the literature, meanwhile decreasing the complexity of the neural network structure. In this work, to verify the robustness and efficiency of our approach, the final proposed dual-path CNN with SVM classifier network is testified utilizing two MRI datasets. The proposed structure achieved an accuracy of 98.3 % on the First dataset, 98.2 % on the Second dataset, and 99.1 % on the combination of the First and the Second datasets. In addition, other assessment measures including Recall, Specificity, Precision, and F1 score are extracted to be 99.5 %, 98.9 %, 97.3 %, and 98.4 %, respectively.