Automatic loose lead detection in sleep EEG data

Authors

• Y Deng
• M Moniruzzaman
• B Rogers
• K Jones
• D Berrigan
• C Matthews
• K Tamura

Abstract

Polysomnography is an overnight sleep study used for diagnosis of sleep disorders. In polysomnography, a patient’s brain activity is measured using electroencephalography (EEG) through six leads placed on the scalp: frontal (F3 and F4), central (C3 and C4), and occipital (O1 and O2). Large artifacts caused by loose leads can distort EEG measurements, but manual checking for such artifacts is prohibitively time-consuming.

We developed a method to automatically identify large artifacts in an EEG trace. After multitaper spectral analysis extracts power for specific frequency bands, we compute, for every 1s time segment, the correlation of band-specific power levels between all pairs of leads. For each lead, we average the pairs involving that lead (e.g., C3-C4, C3-F3, C3-F4, C3-O1, C3-O2 for C3), creating a time series of segment-specific average correlations for each lead. Next, our algorithm scans each time series separately for “bad” segments using a local moving window. A segment is designated “bad” when its correlation is less than half of median correlation among all segments in the window; otherwise, the segment is designated “good”. In a second pass, a segment is declared an outlier and assigned value ‘1’ when its correlation is less than ¼ of the 75th percentile among all “good” segments; otherwise, a segment is assigned value ‘0’.

A continuous period of outliers reveals a loose lead. We scan the temporal sequence of segments by summing outlier values within moving 300s windows and declare a loose lead present within a window when the sum exceeds five.

Scientific Focus Area: Computational Biology

This page was last updated on Monday, September 25, 2023