International Journal of Scientific Research and Engineering Development

One of the most difficult tasks in medical imaging is diagnosing brain tumors, primarily because tumor variability and the complexity of MRI scans. Over the last decade, research has shifted from conventional machine learning models to more advanced deep learning and transfer learning architectures. These approaches have consistently reported promising results, with accuracies ranging from 80% to nearly 98% in classification tasks. At the same time, XAI, or explainable artificial intelligence, has become a critical component, enabling clinicians to visualize and validate automated predictions. Grad-CAM, SHAP, and LIME are several techniques that help close the gap between algorithmic output and medical reasoning, thereby improving trust and clinical usability. This review compiles studies published between 2011 and 2025, examining their methodologies, strengths, and limitations. It also highlights major challenges, including dataset limitations, class imbalance, and lack of integration into real clinical workflows. The review concludes that future systems must strike a balance between predictive accuracy and interpretability to support safe and effective adoption in healthcare practice.