Published Paper Details:

Brain tumors pose a serious threat to health, and early detection and their proper classification is essential for successful treatment. This work proposes an integrated framework, where both classical machine learning and deep learning methods were leveraged to analyze multi-modal MRI scans for tumor localization and classification. We pre-processed MRI data-skull stripping and normalization-followed by feature extraction using both handcrafted descriptors (e.g., GLCM and wavelets) and learned features from deep networks. Advanced segmentation models (e.g., 3D U-Net variants) delineate tumor regions, while object detectors (YOLOv5, Faster R-CNN) rapidly localize tumors in full scans. Lastly, for classification, features are fed into a variety of classifiers and ensemble strategies, including CNNs, SVM, KNN, Random Forest, and Naive Bayes. Extensive experiments on benchmark datasets such as BraTS, Figshare, etc., show our hybrid approach achieves very high accuracy-for instance, a CNN+U-Net model reaches ~98% accuracy on the BraTS segmentation task. Object detectors like YOLOv11 achieve ~99.2% precision/recall on tumor detection. Similarly, the proposed ensemble classifiers show improved classification performance. Quantitative comparisons, as depicted in Table 1, confirm that our methods outperform many existing approaches. These results prove that the integration of different machine learning strategies significantly improves the diagnostic accuracy of brain tumors while maintaining computationally feasible processing, therefore offering a promising tool to assist clinical decision-making.