MRI-Based Brain Tumor Classification Using Transfer Learning

Brain tumors are among critical conditions that deserve explicit diagnosis at the right time. This paper gives an automated method to classify brain tumor using advanced transfer learning methods. In this work, three recent pre-trained convolutional neural networks, ResNet152V2, Xception, and VGG16 have been used for classifying tumors into glioma, meningioma, pituitary tumor, and no tumor. A strong methodology follows the work, carrying out data preprocessing and augmentation with fine-tuning of these architectures so as to adapt them for medical imaging. Among all, the Xception model achieved highest accuracy of 95%, while the ResNet152V2 and VGG16 models follow closely with an accuracy of 92% each. It follows the performance on recall, accuracy, F1-score, and precision metrics. This, therefore, embodies an overview of how transfer learning could further enhance diagnostic accuracy and reliability with the purpose of assisting the clinician in making informed decisions from a range of choices and reducing the chance of diagnostic error in critical situations.