A HYBRID DEEP LEARNING MODEL FOR HIGH-ACCURACY BRAIN TUMOR

Abstract

Brain tumors comprise one of the most dangerous and life-threatening conditions in neurology, and timely and correct diagnosis is of the utmost importance to effective treatment and a positive outcome for patients. The control of artificial intelligence in the medical industry has caused deep learning to become an important tool to address the high-tech need of medical imagery interpretation. The proposed research presents a hybrid deep learning model, combining the VGG-16 and ResNet-50 synergistically with transfer learning in a multi-classification brain tumor Magnetic Resonance Imaging (MRI) scan. The model was rigorously trained and tested on one of the most extensive publicly accessible Brain Tumor MRI datasets that covered a broad multi-facet collection of scans, including glioma, meningioma, pituitary tumors, and non-tumor subjects. All the model training, assessment, and data preprocessing were implemented in Google Colab and were based on TensorFlow and Keras frameworks. The accuracy of the proposed hybrid scheme is 99.1 percent with excellent precision, recall, and F1-score. These results prove the model can be deployed in the clinical decision support system. This hybrid architecture reaffirms the increasing role of deep learning in computer-aided diagnosis (CAD) systems in medicine because it has enhanced the feature learning and classification aspects.