Transfer Learning Based Multiclass Brain Tumor Classification Using MRI Data

Deep learning for classifying brain tumors has transformed medical image analysis, significantly improving upon manual detection techniques. Brain tumors are defined by aber- rant cellular proliferation in the brain or central spinal canal, presenting significant health hazards, including malignant brain cancer. Early detection through Magnetic Resonance Imaging (MRI) is crucial but traditionally labor-intensive and prone to error. This study was conducted to refine the classification of brain tumor images, we introduce new layers to the ResNet50 architecture. Utilizing a comprehensive dataset from SARTAJ, Br35H, and Fig share, we employ data augmentation and pre- processing techniques to improve model training. Our approach of execution is assessed against a number of different deep neural network models, including VGG16, VGG19, EfficientNet- B0, Inception, and DenseNet121, among others. The suggested ResNet50 model, which has been fine-tuned, obtains superior performance, surpassing existing models in F1 score, precision, recall, AUC and ROC, reaching an accuracy level of 99.31%. Advanced deep learning methods can lead to improve the diagnosis accuracy and efficacy of brain tumor, as these findings demonstrate.