Attention-Enhanced EfficientNet Framework for Multi-Class Brain Tumor Classification: A Deep Learning Approach

Classification of brain tumors through Magnetic Resonance Imaging (MRI) is highly crucial for proper diagnosis, treatment plans, and prognosis. MRI offers better contrast between tissues and many types of images, which make it an ideal imaging method for the brain tumor. Nevertheless, manual interpretation is highly time consuming and relies heavily on the skills of the radiologists, with chances of getting different diagnoses. Deep learning algorithms have been proven efficient in automating brain tumor classification tasks. Even though they offer great results, many of the available methods can only handle three classes and have problems handling imbalanced data classes in medical applications. Moreover, the opacity of the Convolutional Neural Networks makes them unreliable when used in healthcare. A scalable deep learning system is proposed that will overcome the imbalances of classes and deliver accurate classification of brain tumor MRIs.