Dementia Etiology Diagnosis via Collaborative Meta Knowledge Enhancement

2026-07-24

AI Summary

This study addresses the significant challenges in etiological diagnosis of dementia disorders such as Alzheimer’s disease, which arise from symptomatic overlap and heterogeneity across multicenter data. To tackle this, the authors propose the Collaborative Meta-knowledge Enhancement (COME) framework, which explicitly encodes metadata—such as acquisition site and imaging modality—into heterogeneity-aware embeddings integrated within a unified Transformer architecture. A trust-region constraint is further introduced to suppress spurious correlations. Evaluated across seven independent cohorts, COME achieves an average macro-AUC of 85.62%, outperforming the strongest baseline by 4.29 percentage points. The method demonstrates exceptional out-of-distribution generalization in cross-center and cross-sequence settings, with its predictions showing strong concordance with amyloid and tau burden as well as clinical severity.

Abstract

Although artificial intelligence (AI) has shown promising performance in several medical tasks, accurate dementia etiology diagnosis with AI remains challenging due to complex overlapping symptoms among diseases. Scaling up the dataset size by combining the cross-center samples may bring a gain in the pursuit of performance, while the inherent data heterogeneity across centers or populations induces the conflict. Conventional multi-task learning paradigms offer a promising framework; however, they fail to consider critical meta information (e.g., site-specific acquisition and modality availability) to combat the heterogeneity. To address this challenge, we propose a Collaborative Meta Knowledge Enhancement (COME) framework for dementia etiology diagnosis, which injects multi-center acquisition semantics, source identifiers, and modality indicators as heterogeneity-aware embeddings into a unified Transformer architecture for scale-up training, enabling explicit modeling of heterogeneity. Besides, a trust-region constrained optimization scheme is designed to regularize the model from spurious correlations during training through a reference model. Across seven independent cohorts, our method achieves state-of-the-art in-domain performance with a mean macro-averaged AUC of 85.62% and a 4.29-point gain over the strongest baseline, while maintaining superior out-of-domain generalization under both cross-center and cross-sequence evaluations. Extensive validation also confirms the alignment between model predictions and established biomarkers (amyloid, tau) and clinical severity, highlighting the potential of COME to enable robust and interpretable dementia diagnostics in real-world settings.

Problem

• dementia etiology diagnosis
• data heterogeneity
• multi-center data
• symptom overlap
• AI in medical diagnosis

Innovation

• Collaborative Meta Knowledge Enhancement
• heterogeneity-aware embedding
• multi-center learning
• trust-region constrained optimization
• Transformer-based diagnosis