Optimized VGG16-Based Deep Learning Framework for Robust Multi-Class Brain Tumor Classification Using MRI Images

The diagnosis of brain tumors through magnetic resonance imaging (MRI) is a delicate and demanding undertaking since the structures of the various tumors have complex variations and visual similarities. Interpretation of MRI scans manually is both time consuming and highly subjective based on the experience of the radiologist thus the reason behind the necessity to develop reliable automated diagnostic systems. The use of deep learning models, especially residual networks, has shown good results in brain tumor classification in recent years. Nevertheless, classical convolutional neural network-based structures, including the VGG family, have not been sufficiently studied in modern medical imaging studies. This paper will suggest an efficient VGG16-based deep learning model to classify brain tumors based on MRI images and with multi-classes accuracy. The suggested method sub-categorizes the MRI scans as glioma, meningioma, pituitary tumor, and no tumor. ImageNet pretrained VGG16 is used as a feature extractor to perform transfer learning and then a customized classification head with regularization methods to enhance the generalization factor. Extensive preprocessing and data augmentation schemes are implemented to provide superior resistance to changes in imaging conditions.