Comparative Study of Deep Learning Architectures for Brain Tumor Classification

Brain tumor classification is an important medical imaging problem because early and correct diagnosis greatly enhances patient survival. In this paper, we investigate automatic brain tumor classification from MRI with deep learning models. We try different pre-trained CNN models, i.e., VGG19, ResNet101, EfficientNetB0, MobileNetV2, InceptionV3, and DenseNet121, on a Kaggle dataset of 13,196 MRI scans into four categories: pituitary, glioma, meningioma, and no tumor. EfficientNetB0 performed best among the other models with a highest accuracy rate of 98.63%. Data augmentation and transfer learning techniques were employed to train the model such that it stabilizes the performance. This paper describes an efficient, automated, and accurate method for detection of brain tumors, establishing the capability of deep learning in clinical diagnostics. The research proves the capability of AI medical imaging software to allow physicians to diagnose patients quicker and treat them better