NeuroDyn-EEG: An Interpretable Pre-trained Model for EEG Based on Neural Dynamics

2026-09-29

AI Summary

This study addresses the lack of anatomical and physiological interpretability in deep learning models for clinical EEG analysis by proposing NeuroDyn-EEG, a novel pretraining framework. This approach is the first to couple an extended Jansen-Rit neural mass model with source projection and parameter inversion, integrated with lightweight pretraining techniques. It achieves precise mapping from scalp EEG signals to regional brain dynamic parameters, thereby bridging representation learning and mechanistic neurophysiology. The proposed framework attains state-of-the-art classification performance across multiple clinical benchmark tasks while successfully revealing disease-specific alterations in neural dynamics. These results demonstrate strong robustness and substantial potential for clinical application.

Abstract

Clinical scalp electroencephalography (EEG) offers a noninvasive window into neural dynamics of neuropsychiatric disorders. However, discriminative deep models often lack anatomically indexed physiological interpretability. We propose NeuroDyn-EEG, a pretraining framework integrating generative priors from neural dynamics. It couples an extended Jansen-Rit neural mass model, leadfield-based source projection, and simulation-based parameter inversion. Trained on synthetic parameter-EEG pairs within physiological ranges, NeuroDyn-EEG estimates 11 regional parameter families across 90 AAL regions plus one global parameter from standard 19-channel EEG, using only ~2.43M trainable parameters. We evaluate the framework across three levels. First, controlled simulations demonstrate robust parameter recovery under diverse noise conditions, while real resting-state EEG evaluations confirm spectral and phase consistency in an inverse-forward closed loop. Second, on four clinical benchmarks (AD65, PD31, Figshare MDD, and TUAB), NeuroDyn-EEG achieves competitive classification performance, securing the highest BACC, AUROC, and AUCPR on PD31 and MDD, and highest BACC on AD65. Third, post hoc regional analyses reveal disease-specific alterations: local synaptic connectivity C_1 involves the most altered regions in AD65, whereas the firing threshold theta ranks first in MDD, offering testable mechanistic hypotheses. Overall, NeuroDyn-EEG maps scalp EEG to anatomically indexed dynamical parameters, bridging representation learning and mechanistic neurophysiology. Code: https://github.com/Gnosis-Neurodynamics/NeuroDyn-EEG.

Problem

• EEG
• interpretability
• neural dynamics
• deep learning
• neuropsychiatric disorders

Innovation

• Neural mass model
• EEG pretraining
• Parameter inversion
• Physiological interpretability
• Neural dynamics