Brain Tumor Classification Using Deep Transfer Learning CNN Models

Authors:

Abstract
Brain tumors are especially prevalent in children and older people. It is severe cancer characterized by the uncontrolled development of brain cells within the skull. It is notoriously difficult to distinguish tumor cells due to their variety. Data augmentation and an image data generator were used to enhance the contrast of brain tumor MRI image cells. The benefit of using a deep CNN model is that it can accurately classify a small image database. Deep features were extracted using the CNN models EfficientNetB2, EfficientNetB3, EfficientNetB5, and InceptionResNetV2. The classification precision measures the efficacy of this research. EfficientNetB2 had an accuracy of 96%, EfficientNetB3 had an accuracy of 97%, and EfficientNetB5 had an accuracy of 96.23%. Among these models, InceptionResNetV2 has provided the most high level of accuracy. As a result, the proposed technique achieves exceptional overall performance. The experiment's findings demonstrate that the proposed model was 99.69% accurate during training and 97.50% accurate during testing. Thus, the proposed deep-learning model can assist in the earlier identification of brain tumors by physicians and radiologists.