Advanced Hybrid AI Methods for Brain Tumor Analysis: Detection and Categorization

This study uses an MRI, or magnetic resonance imaging a non-invasive method that produces high-contrast greyscale pictures, to accurately classify brain cancers, specifically pituitary, glioma, and meningioma. Convolutional neural networks (CNNs), which offer notable improvements in diagnosis accuracy, have become essential in medical imaging with the emergence of deep learning. This study uses five pre-trained CNN models using transfer learning (TL) on the publically available CE-MRI (Figshare) dataset. For tumor classification, the process entails incorporating bespoke layers, among them fully connected (FC) layers, and initializing ResNets with ImageNet weights. Extensive tests that use data augmentation to improve classification accuracy assess the resilience of these optimized ResNets. Grad-CAM visualization provides interpretation in tumor localization by highlighting the model's focal points. With an Accuracy of 95.72%, Sensitivity of 95.89%, Specificity of 96.98%, Precision of 94.05%, and an F1Score of 95.01%, ResNet50 outperforms the other examined designs in terms of operational efficiency and performance. These findings demonstrate how well this method works to increase the diagnostic precision of brain tumor classification.