Deep Jansen-Rit Parameter Inference for Model-Driven Analysis of Brain Activity

TL;DR

This work tackles the challenge of inferring JR-NMM parameters from EEG for model-driven EC analysis. ItBenchmarking four inverse-modeling approaches—EEGTransformer, Vanilla LSTM, CNN-BiLSTM, and SBI-based SNPE—across noise conditions reveals that local parameters such as $A_e$, $A_i$, $b_e$, and $b_i$ are recoverable, while connectivity parameters $a_1$-$a_4$ are not reliably identifiable from ERP alone. Transformer-based architectures achieve the strongest, most noise-robust recoveries (notably for $b_i$ with correlations >0.9), and sensitivity analyses align parameter influence with ERP morphology, guiding identifiability understanding. The findings provide a scalable framework for whole-brain EC inference from EEG and motivate future work with empirical data, transfer learning, and multi-region modeling to extend applicability.

Abstract

Accurately modeling effective connectivity (EC) is critical for understanding how the brain processes and integrates sensory information. Yet, it remains a formidable challenge due to complex neural dynamics and noisy measurements such as those obtained from the electroencephalogram (EEG). Model-driven EC infers local (within a brain region) and global (between brain regions) EC parameters by fitting a generative model of neural activity onto experimental data. This approach offers a promising route for various applications, including investigating neurodevelopmental disorders. However, current approaches fail to scale to whole-brain analyses and are highly noise-sensitive. In this work, we employ three deep-learning architectures--a transformer, a long short-term memory (LSTM) network, and a convolutional neural network and bidirectional LSTM (CNN-BiLSTM) network--for inverse modeling and compare their performance with simulation-based inference in estimating the Jansen-Rit neural mass model (JR-NMM) parameters from simulated EEG data under various noise conditions. We demonstrate a reliable estimation of key local parameters, such as synaptic gains and time constants. However, other parameters like local JR-NMM connectivity cannot be evaluated reliably from evoked-related potentials (ERP). We also conduct a sensitivity analysis to characterize the influence of JR-NMM parameters on ERP and evaluate their learnability. Our results show the feasibility of deep-learning approaches to estimate the subset of learnable JR-NMM parameters.

Figures (4)

• Figure 1: Flow diagram for the inference of $A_e$, $A_i$, $b_e$, $b_i$, $a_1$-$a_4$, and $C$ using the EEGTransformer, LSTM, CNN-BiLSTM, and SBI methods. The JR model generates simulated EEG for training and testing.
• Figure 2: Sensitivity analysis for synaptic parameters ($A_e$, $A_i$, $b_e$, $b_i$) showing the ERP amplitude (left), error (middle), and gradient (right).
• Figure 3: Sensitivity analysis for local connectivity ($a_1$-$a_4$) and C showing the ERP amplitude (left), error (middle), and gradient (right).
• Figure 4: Comparison of deep learning models estimating JR-NMM parameters across noise levels. EEGTransformer and CNN-BiLSTM maintaining high correlations for $b_i$. The vanilla LSTM consistently underperforms with lower correlations, while SBI exhibits parameter-dependent variability.