Canonical Correlation Analysis and Multi-Channel Cardiography Improve Artefact Cleaning in Heartbeat-Locked Analyses

Abstract

Cardiac artefacts are problematic for neurophysiological analyses, especially for heartbeat-locked brain responses where neural activity and artefacts co-occur in time. Traditional approaches using single-channel electrocardiography for artefact removal do not fully capture the multidimensional spread of cardiac fields. Moreover, even if several channels are recorded, no established methods exists for integrating them for artefact removal. Here, we propose a multivariate approach based on Canonical Correlation for cardiac artefact removal and report its effectiveness in electroencephalography recorded simultaneously with a custom 15-channel electrocardiography in 14 participants. We quantify cleaning quality via residual R-peak artefact and neural signal preservation via alpha power. Canonical Correlation systematically outperforms the traditional Independent Component Analysis in both reducing R-peak artefacts and preserving alpha power. By analysing all possible channel combinations, we found that neck or supraclavicular electrodes improve cleaning when only a few channels are available. With four or five channels, precordial electrodes in combination with limb or supraclavicular locations provide a performance comparable to the 15-channel setup. While these findings require validation in other datasets, we outline clear decision criteria for cleaning efficiency and show that canonical correlation is a reliable approach for multi-channel cardiac artefact removal.