EEG Bad-Channel Detection Using Multi-Feature Thresholding and Co-occurrence of High-Amplitude Transients

Abstract

Bad channels in electroencephalography (EEG) recordings can substantially degrade downstream analyses, particularly in high-density datasets where localized hardware or motionrelated artifacts may affect groups of electrodes in a structured manner. We introduce a MATLAB Module for bad-channel quality control that emphasizes interpretability, relational structure, and human-in-the-loop validation rather than fully automated rejection. The method operates on multichannel EEG data and combines complementary channel-level features, including time-dependent neighbor dissimilarity and amplitude- and variance-based statistics to score and pre-label channels as good, suspicious, or bad. To expose shared artifactual structure, channels are additionally grouped using a similarity measure derived from the co-occurrence of robustly detected high-amplitude transients, allowing channels to be reviewed together. Importantly, clustering is used as an exploratory tool to reveal co-artifactual patterns rather than to impose final class labels, which are confirmed through an interactive review interface supported by summary visualizations and grouped channel displays. This Module is released as a publicly available codebase with documentation, example work-flows, and a supporting dataset. This Module is designed as a quality-control step preceding ICA and does not replace end-to-end data cleaning pipelines, which typically involve additional steps such as interpolation of known bad channels.