DW-DGAT: Dynamically Weighted Dual Graph Attention Network for Neurodegenerative Disease Diagnosis

TL;DR

The paper tackles early diagnosis of the neurodegenerative diseases Parkinson's and Alzheimer's by fusing multi-metric neuroimaging (1D/2D/3D) and phenotypic data within a dual graph attention framework. It introduces DW-DGAT, which combines a data fusion module, a single-ROI graph attention (SGA) for micro-level features, a global graph attention (GGA) for macro-level sample relationships, and a class weight generator (CWG) to address class imbalance with stable losses. Key contributions include a general-purpose multi-form data fusion strategy, a MHSA-based dual-graph attention architecture, and a dynamic weighting mechanism that yields state-of-the-art results on PPMI and ADNI3 datasets, along with extensive ablations and complexity analyses. The approach improves robustness to data heterogeneity and imbalance, offering a scalable solution with practical implications for clinical ND diagnosis and monitoring.

Abstract

Parkinson's disease (PD) and Alzheimer's disease (AD) are the two most prevalent and incurable neurodegenerative diseases (NDs) worldwide, for which early diagnosis is critical to delay their progression. However, the high dimensionality of multi-metric data with diverse structural forms, the heterogeneity of neuroimaging and phenotypic data, and class imbalance collectively pose significant challenges to early ND diagnosis. To address these challenges, we propose a dynamically weighted dual graph attention network (DW-DGAT) that integrates: (1) a general-purpose data fusion strategy to merge three structural forms of multi-metric data; (2) a dual graph attention architecture based on brain regions and inter-sample relationships to extract both micro- and macro-level features; and (3) a class weight generation mechanism combined with two stable and effective loss functions to mitigate class imbalance. Rigorous experiments, based on the Parkinson Progression Marker Initiative (PPMI) and Alzhermer's Disease Neuroimaging Initiative (ADNI) studies, demonstrate the state-of-the-art performance of our approach.

Figures (10)

• Figure 1: The architecture of the proposed methodology.
• Figure 2: Loss histories of RA-GCN and DW-DGAT.
• Figure 3: ROC curves of all networks.
• Figure 4: The t-SNE visualization results of three networks on two datasets, where class-0 in purple indicates normal.
• Figure A1: PANDA tracking options.