Brain Tumour Three-Class Classification on MRI Scans Using Transfer Learning and Data Augmentation

Abstract

Accurate classification is a prerequisite for brain tumour diagnosis. The proposed method is a modified computer-aided detection (CAD) technique used for leveraging automatic classification in brain magnetic resonance imaging (MRI), where the model has trained a pipeline of convolutional neural networks (CNNs) using transfer learning (TL) on ResNet 50 with PyTorch. The proposed method employs benchmarked datasets from figshare database, where data augmentation (DA) is applied to increase the number of datasets that can further increase the training efficiency. Thus, the retrained model can classify the tumour images into three classes, i.e., glioma, meningioma, and pituitary tumours. Classification accuracy was tested by comparing the accuracy matrices, loss matrices, and confusion matrix and found to be 99%. The proposed model is the first of its kind that employs both DA and TL on the ResNet 50 model for performing a three-class classification on brain tumour, and results reveal that it outperforms all other existing methods.KeywordsComputer-aided detectionMagnetic resonance ımagingConvolutional neural networkTransfer learningResNet 50Data augmentation