MRI-Based Brain Tumor Detection Using Transfer Learning: Comparative Analysis of CNN Models and a Hybrid Ensemble Approach

Early and precise identification of brain tumors is essential for improving diagnostic accuracy and supporting timely medical intervention. The analysis of MRIs manually is quite time-consuming and demands expertise. To mitigate the previously identified challenges, an intelligent machine learning approach is developed for accurate brain tumor detection and classification from MRI scans. For this purpose, a set of convolutional neural networks such as ResNet50, DenseNet121, Xception, EfficientNetV2B0 has been utilized. These deep learning models employ transfer learning to extract meaningful features from medical images. Each model was trained separately and evaluated depending on accuracy, precision, recall, f1-score, ROC-AUC. The performance of all four models is compared to determine their strengths and weaknesses. Furthermore, we tried merging the results obtained from each model. Based on the obtained results, the ensemble approach demonstrates superior performance compared to standalone models.