Bridging Neural and Hemodynamic Pathways through EEG–bbNIRS Fusion for Early Alzheimer's Disease Classification

This paper presents an adaptive multimodal fusion network that integrates EEG and broadband near-infrared spectroscopy (bbNIRS) signals for early Alzheimer's disease (AD) classification. By bridging complementary neural and hemodynamic pathways, the framework captures temporal electrophysiological dynamics alongside metabolic and vascular biomarkers. We stabilize training through targeted data augmentations, adaptive regularization, and learning-rate scheduling. Modality contribution analyses highlight the complementary nature of EEG and bbNIRS across AD progression stages. The proposed method achieves 7–12% improvement in AUC over unimodal baselines with robust cross-validation, demonstrating its potential as a non-invasive, cost-effective tool for early AD screening.