A Systematic Review of Advances in Brain Disease Detection using Convolutional Neural Networks and Explainable Artificial Intelligence Techniques

Accurate and interpretable tumor classification remains a critical challenge in medical image analysis. In this study, we conduct a comprehensive evaluation of ten state-of-the-art convolutional neural network (CNN) architectures, including InceptionV3, Xception, MobileNetV2, DenseNet121, NASNetMobile, VGG16, VGG19, ResNet50, ResNet101, and EfficientNetB0, on a curated dataset of tumorous and nontumorous images. Each model’s performance was rigorously assessed using standard classification metrics: accuracy, precision, recall, and F1-score. InceptionV3 emerged as the top- performing model with an accuracy of 97.75%, while EfficientNetB0 showed the lowest at 56.50%. Beyond raw performance, we prioritized model transparency by applying five explainable AI (XAI) methods—Grad-CAM, Saliency Maps, Integrated Gradients, Vanilla Gradients, and SmoothGrad—to visualize and interpret the models’ decision-making processes. These visualizations revealed critical insights into model attention and class-specific feature relevance, reinforcing the importance of explainability in medical diagnostics. The results not only highlight the superiority of modern CNNs in medical imaging tasks but also emphasize the value of interpretability tools for building trust and accountability in clinical AI applications.