An Interpretable Deep Learning Framework For Mri-based Alzheimer’s Disease Detection And Automated Clinical Reporting

Abstract

Alzheimer’s Disease (AD) Is A Progressive Neurodegenerative Disorder That Affects Memory, Cognitive Abilities, And Overall Brain Function, Particularly In The Aging Population. Early And Reliable Diagnosis Is Essential For Timely Intervention, Disease Management, And Improved Patient Care. Magnetic Resonance Imaging (MRI) Provides Non-invasive Structural Information About The Brain And Has Become An Important Source Of Data For Computer-aided Alzheimer’s Disease Assessment. Although Deep Learning Techniques Have Demonstrated Considerable Potential For MRI-based Disease Classification, Many Existing Approaches Primarily Emphasize Predictive Performance While Providing Limited Insight Into The Reasoning Behind Their Decisions. Furthermore, Conventional Diagnostic Models Generally Do Not Provide An Integrated Mechanism For Transforming Model Predictions Into Understandable Clinical Information. To Address These Limitations, This Paper Proposes An Interpretable Hybrid Deep Learning And Generative Artificial Intelligence (AI) Framework For MRI-based Multi-stage Alzheimer’s Disease Diagnosis With Automated Clinical Report Generation. The Proposed Framework Combines A Convolutional Neural Network (CNN) And A Swin Transformer To Learn Complementary Local Structural And Global Contextual Representations From MRI Brain Images. The Extracted Features Are Fused For Multi-stage Classification Of Normal, Mild, And Moderate Conditions. Grad-CAM Is Incorporated To Identify And Visualize The Brain Regions That Contribute To The Classification Decision, Thereby Improving Model Transparency. Subsequently, A Generative AI-based Large Language Model (LLM) Converts The Verified Classification Result, Confidence Information, And Visual Explanation Into A Human-readable Clinical-style Report. Unlike Prediction-only Systems, The Proposed Framework Establishes A Unified Pipeline Connecting Diagnosis, Visual Explanation, And Evidence-guided Report Generation. The Framework Is Intended To Improve Diagnostic Interpretability, Communication, And Usability While Providing A Foundation For Trustworthy AI-assisted Decision Support In Alzheimer’s Disease Assessment.

Keywords