Multimodal AI for Biomarker and Etiological Assessment of Dementias

TL;DR

This paper introduces an AI system that combines multiple data types to better diagnose dementia and assess biomarkers like amyloid beta and tau.

Contribution

The novel contribution is a multimodal AI framework for scalable dementia screening and biomarker estimation without requiring advanced imaging.

Findings

The framework can identify dementia causes and estimate amyloid beta/tau PET status using diverse data.

It handles mixed dementia cases and incomplete data, enabling personalized dementia management.

The approach supports scalable screening for Alzheimer's therapies and clinical trials.

Abstract

Dementia differential diagnosis and biomarker assessment pose significant challenges due to symptom overlap and limited access to advanced imaging like PET for detecting amyloid beta (Aβ) and tau (τ) in neurodegenerative conditions such as Alzheimer’s disease (AD). In this talk, I will highlight our lab’s recent work on multimodal AI frameworks that fuse diverse data to identify dementia etiologies, handle mixed dementias and incomplete cases, and estimate Aβ/τ PET status. This approach promises scalable screening for personalized dementia management, AD therapies, and clinical trials.