MontageFlow produces realistic EEG signals that follow physical rules
MontageFlow: Physically Valid High Fidelity EEG Generation
Artificial Intelligence
Summary
EEG data, which records brain signals from multiple electrodes on the scalp, must follow strict physical relationships between channels. Many current computer models that try to create synthetic EEG data ignore these rules, leading to unrealistic results. This paper presents MontageFlow, a new method that generates EEG signals while strictly preserving these physical channel relationships. This method also better captures important brain activity features, producing more accurate and realistic synthetic EEG data than previous approaches.
What this means in practice
- •For clinical data teams: Generate realistic synthetic EEG data that obey physical electrode relationships for training and testing diagnostic tools under privacy constraints.
- •For neuromarketing analysts: Create high-fidelity synthetic EEG signals to simulate brain responses for market research without needing sensitive participant data.
Authors
Sudarshan Regmi, Nabaraj Subedi, Arvind Pillai, Yu Yvonne Wu, Yuliang Chen, Michael V. Heinz, Lisa Marsch, Nicholas C. Jacobson, Andrew Campbell
Abstract
Generative modeling offers a promising solution to the scarcity of labeled clinical electroencephalography (EEG) data under strict privacy constraints. Although recent generative models report substantial improvements in generation fidelity, we show that these gains conceal a categorical failure. EEG is not an arbitrary multichannel signal: each channel records the voltage difference between two scalp electrodes. So, real recordings satisfy exact linear identities among their channels. Current generators invariably violate these identities, yet this has gone unnoticed. We therefore propose MontageFlow, a flow-matching model that enforces these identities by restricting its output to signals that satisfy them. We find that ensuring validity improves fidelity too. Furthermore, to retain the EEG characteristics that oversmoothing washes out, we also condition the model on a physiological state vector encoded from the whole EEG window. For generation, this vector is drawn from a prior fitted in latent space. Across 3 large-scale benchmarks, our model improves TS-FID by 1.6-3.6$\times$ over the closest rival while ensuring validity.