petVAE: A Data-Driven Model for Identifying Amyloid PET Subgroups Across the Alzheimer's Disease Continuum

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Amyloid-β (Aβ) PET imaging is crucial for Alzheimer's disease (AD) diagnosis.
• Current classification is often binary (Aβ+/Aβ-), limiting granularity.
• Subtle Aβ accumulation patterns may indicate early disease stages.

Purpose of the Study:

• To develop a model for identifying subgroups along the Aβ accumulation continuum.
• To analyze Aβ PET scans beyond binary classification.
• To establish a data-driven framework for studying AD progression.

Main Methods:

• Utilized 3,110 Aβ PET scans from ADNI and A4 datasets.
• Developed and applied petVAE, a 2D variational autoencoder, for scan reconstruction and feature extraction.
• Clustered scans based on extracted latent features to identify subgroups.

Main Results:

• petVAE accurately reconstructed Aβ PET scans.
• Clustering identified four distinct subgroups (Aβ-, Aβ-+, Aβ+, Aβ++).
• Clusters differed significantly in biomarker levels (CSF Aβ, tau), APOE ε4 status, cognitive performance, and AD progression rates.

Conclusions:

• petVAE effectively captures the Aβ continuum in PET scans.
• The identified clusters represent biologically meaningful stages of AD.
• This approach enhances early preclinical AD detection and disease progression studies.