Convolutional Monge Mapping between EEG Datasets to Support Independent Component Labeling

TL;DR

EEG spectral differences across recording setups can degrade IC labeling. The authors extend CMMN with channel-averaged PSDs and subject-to-subject mapping to enable domain adaptation when datasets have different numbers of channels or hardware configurations. They apply CMMN to pre-trained IC classifiers using PSD and autocorrelation features, evaluating on Emotion (US) as source and Cue (Europe) as target, and show brain IC labeling performance exceeding ICLabel on the target dataset. The work demonstrates practical improvements in artifact removal and automatic IC labeling, supporting more robust EEG analyses across diverse recording environments.

Abstract

EEG recordings contain rich information about neural activity but are subject to artifacts, noise, and superficial differences due to sensors, amplifiers, and filtering. Independent component analysis and automatic labeling of independent components (ICs) enable artifact removal in EEG pipelines. Convolutional Monge Mapping Normalization (CMMN) is a recent tool used to achieve spectral conformity of EEG signals, which was shown to improve deep neural network approaches for sleep staging. Here we propose a novel extension of the CMMN method with two alternative approaches to computing the source reference spectrum the target signals are mapped to: (1) channel-averaged and $l_1$-normalized barycenter, and (2) a subject-to-subject mapping that finds the source subject with the closest spectrum to the target subject. Notably, our extension yields space-time separable filters that can be used to map between datasets with different numbers of EEG channels. We apply these filters in an IC classification task, and show significant improvement in recognizing brain versus non-brain ICs. Clinical relevance - EEG recordings are used in the diagnosis and monitoring of multiple neuropathologies, including epilepsy and psychosis. While EEG analysis can benefit from automating artifact removal through independent component analysis and labeling, differences in recording equipment and context (the presence of noise from electrical wiring and other devices) may impact the performance of machine learning models, but these differences can be minimized by appropriate spectral normalization through filtering.

Figures (3)

• Figure 1: Frequency response of filters learned for mapping the Cue dataset to the Emotion dataset. (Left) Barycenter mapping scheme with a $\ell_1$-normalized barycenter. Notice that the noise at 50 Hz is reduced and the noise at 60 Hz is relatively amplified. Overall, the filter attenuates since the normalized barycenter has lower magnitudes.(Right) Subj-to-subj mapping scheme. Even though the mapping scheme is different, the line noises are visibly still being 'swapped'.
• Figure 2: (Top) Channel-averaged power spectral densities for each subject in the Emotion dataset. Notice that some subjects are outliers in terms of overall amplitude. (Bottom) This is the $\ell_1$-normalized barycenter computed from the source Emotion dataset. In the Barycenter mapping scheme, all target signals are filtered such that their robust channel-average PSD matches this. Notice the large spike at 60 Hz.
• Figure 3: PSDs for different subjects in the target Cue dataset. Notice the spike at 50 Hz.