An Explainable Web-Based Diagnostic System for Alzheimer's Disease Using XRAI and Deep Learning on Brain MRI.

Area of Science:

• Artificial Intelligence in Medicine
• Neuroimaging Analysis
• Machine Learning for Disease Classification

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder characterized by cognitive decline.
• Current AI-driven neuroimaging for AD lacks clinical interpretability and usability.
• Explainable AI (XAI) frameworks, like XRAI, can enhance clinical decision-making through visualizations.

Purpose of the Study:

• To develop and evaluate a clinically deployable AI system for Alzheimer's disease severity classification using 2D brain MRI.
• To integrate XRAI for enhanced interpretability of AI model predictions.
• To create a user-friendly web interface for real-time AD diagnosis support.

Main Methods:

• Trained three deep learning models (MobileNet-V3 Large, EfficientNet-B4, ResNet-50) on an augmented Kaggle MRI dataset (33,984 images) across four AD severity classes.
• Evaluated model performance on both augmented and original datasets, incorporating XRAI for region-based attribution mapping.
• Developed a Gradio-based web interface for real-time predictions and visual explanations.

Main Results:

• MobileNet-V3 demonstrated superior performance with high accuracy (99.18% augmented, 99.47% original) and minimal parameters (4.2 M).
• XRAI visualizations correlated with known AD neuroanatomical patterns, improving clinical interpretability.
• The web interface provided sub-20-second inference times with high confidence, supporting clinical workflows.

Conclusions:

• This research marks the first systematic integration of XRAI for MRI-based deep learning AD severity classification.
• The MobileNet-V3 system offers a practical solution with high accuracy, efficiency, and interpretability for clinical use.
• The study presents a viable pathway for adopting explainable AI in early and accurate Alzheimer's disease detection.