Augmented MRI Images for Classification of Normal and Tumors Brain through Transfer Learning Techniques

Abstract

A brain tumor is a severe malignant condition caused by uncontrolled and abnormal cell division. Recent advances in deep learning have aided the health business in Medical Imaging for the diagnosis of numerous disorders. The most frequent and widely used deep learning algorithm for visual learning and image recognition. This research seeks to multi-classification tumors in the brain from images attained by Magnetic Resonance Imaging (MRI) using deep learning models that have been pre-trained for transfer learning. As per the publicly available MRI brain tumor dataset, brain tumors identified as glioma, meningioma, and pituitary, are accounting for most brain tumors. To ensure the robustness of the suggested method, data acquisition, and preprocessing are performed in the first step followed by data augmentation. Finally, Transfer Learning algorithms including DenseNet, ResNetV2, and InceptionResNetv2 have been applied to find out the optimum algorithm based on various parameters including accuracy, precision, and recall, and are under the curve (AUC). The experimental outcomes show that the model’s validation accuracy is high for DenseNet (about 97%), while ResNetv2 and InceptionResNetv2 achieved 77% and 80% only.