Leveraging Early Diagnosis of Alzheimer's Disease Using Deep Learning and XAI

This paper presents a high-performance and interpretable deep learning system for early diagnosis and stage classification of Alzheimer's disease (AD) using brain MRI data. Using the clinically validated ADNI dataset, which was supplemented with a numerical Alzheimer's dataset from Kaggle, the proposed models achieved remarkable accuracy: 99.14 % with CNN+K-Nearest Neighbors (KNN), 98.12% with standalone CNN, 97.34 % with AdaBoost, and 91.33 % with Random Forest. To address the severe class disparity, particularly in the “Moderate Demented” category, sophisticated data augmentation and resampling techniques were employed. Explainable AI (XAI) tools like Grad-CAM, SHAP, and LIME were used to display and quantify the brain regions and traits that most affected the designed models' predictions. These results demonstrate the technical soundness and high accuracy of the proposed AI system. Future studies could look into expanding multi-modal data (like genetic and cognitive tests), integrating with clinical decision support systems, and deploying advanced ensemble CNN architectures (like ResNet, VGG16, and Inception) to enhance generalization and real-world scalability.