Brain Tumor Classification and Segmentation Using Transfer Learning from MRI Images

Abstract

Manual diagnosis of tumors on magnetic resonance images (MRIs) entails more time and doing so increases the risk of human error and incorrect tumor type identification and classification. Cells develop quickly and uncontrollably, which causes brain tumors and may cause death unless handled in the beginning stages. Therefore, a transfer learning framework for brain tumor classification is presented to simplify the task of healthcare professionals by automating tough medical processes. MRI image analysis was done on a publicly available dataset from Kaggle. The proposed method is implemented on VGG16, ResNet50, EfficientNetB0, and U-Net architectures. Training accuracy and test accuracy of all these four neural architectures compared and results that U-Net performs better in the classification of brain tumors compared to the other networks.