SMAS: Structural MRI-based AD Score using Bayesian supervised VAE

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Biomarker Development

Background:

• Alzheimer's Disease (AD) poses a significant challenge, necessitating advanced diagnostic tools.
• Current methods for assessing AD-related brain changes can be limited in sensitivity and interpretability.
• Structural MRI is a key modality for observing neurodegeneration in AD.

Purpose of the Study:

• To introduce and validate the Structural MRI-based Alzheimer's Disease Score (SMAS), a novel deep learning index.
• To assess SMAS's association with cognitive function, age, and brain morphometry.
• To evaluate SMAS's utility in early AD detection and longitudinal monitoring.

Main Methods:

• Development of a deep learning Bayesian-supervised Variational Autoencoder (Bayesian-SVAE) to create the SMAS index.
• Utilized baseline structural MRI data from the DELCODE cohort.
• Longitudinal validation in independent DELCODE and Alzheimer's Disease Neuroimaging Initiative (ADNI) cohorts.

Main Results:

• SMAS demonstrated strong correlations with cognitive performance, age, hippocampal, and gray matter volumes across cohorts.
• The SMAS index showed high accuracy in distinguishing healthy individuals from those with AD (AUC up to 0.971).
• SMAS outperformed existing biomarkers (SPARE-AD, hippocampal volume) in longitudinal tracking over 36 months.

Conclusions:

• SMAS is a sensitive and interpretable biomarker reflecting AD-related brain atrophy.
• SMAS shows significant potential for early AD detection and monitoring disease progression.
• The relevance map analysis highlights SMAS's focus on key AD-affected brain regions.