Interpretable MRI-Based Deep Learning for Alzheimer's Risk and Progression.

Area of Science:

• Neuroimaging and Artificial Intelligence
• Biomarker Discovery for Neurodegenerative Diseases

Background:

• Early detection of Alzheimer's disease (AD) is crucial for timely intervention, especially with emerging immunotherapies.
• There is a need for accessible and efficient biomarkers for early AD diagnosis.

Purpose of the Study:

• To evaluate the cross-ethnic generalization and clinical utility of a previously developed MRI-based deep learning model for AD detection.
• To assess the model's ability to predict future AD progression and identify AD subtypes.

Main Methods:

• Applied a pre-trained deep learning model, developed using North American data, to a large Chinese cohort (SILCODE) with 1,105 brain MRI scans from 722 participants.
• Utilized an interpretable deep learning brain risk map approach to identify AD subtypes.
• Correlated model-derived risk scores with cognitive assessments and plasma biomarkers (tau proteins, NfL).

Main Results:

• The MRI-based deep learning model demonstrated robust cross-ethnic generalization without retraining, achieving 91.3% AUC and 95.2% sensitivity for AD classification.
• The model accurately identified 86.7% of individuals at risk of AD progression over 5 years, with high-risk individuals showing faster progression.
• Identified distinct AD brain subtypes, including a Mild Cognitive Impairment (MCI) subtype linked to rapid decline, and risk scores correlated significantly with cognitive and plasma biomarkers.

Conclusions:

• MRI-based deep learning models exhibit strong generalizability and clinical utility in diverse populations for early AD detection and risk stratification.
• The model provides valuable tools for early therapeutic intervention by identifying at-risk individuals and subtypes.
• The open-source model and free online tool facilitate widespread early screening and intervention for Alzheimer's disease.