An automated pontine-wave detection system.

Abstract

Pontine-waves (P-waves), the pontine component of ponto-geniculo-occipital waves, represent a close marker of brainstem phasic events and are associated with cardiorespiratory changes in sleep. Because visual scoring is subjective and cumbersome, we developed an automated P-wave analysis system, which could detect and classify P-waves as clusters or as isolated events in bipolar recordings of the pontine electroencephalogram (EEG) of conscious rats. A computer algorithm was developed to extract and normalize each half-wave of the pontine EEG according to the background noise level. Candidate events for different P-wave patterns (uniphasic, biphasic, and triphasic) were compared to a corresponding set of amplitude and duration thresholds to identify P-waves and to reject artifacts. Ten adult male Sprague-Dawley rats were instrumented for chronic polysomnography. Two human experts manually scored each recording, and their consensus score was used as the "gold standard" for algorithm optimization and validation. The algorithm's scoring thresholds were optimized on a training set of 5 six-hour polysomnographic records, yielding 96.8% accuracy and 97.7% sensitivity versus human consensus scoring. Validation of the algorithm, using the optimized threshold values, was conducted using a set of 5 independent recordings, resulting in 94.8% accuracy and 94.7% sensitivity versus human consensus scoring. We have developed and validated an automated system for detection and classification of P-waves in conscious rats with advantages over human scoring, including increased speed and perfect reliability.