Item

Abstract

Physiological research and evaluation of tools for automated sleep analysis either require resource-demanding data acquisition or rely on open-access databases. Publicly available polysomnographic data feature poor spatiotemporal resolution, usually less than 10 electroencephalography
(EEG) channels and sampling rates below 250 Hz; therefore, are unsuitable for analyses with high resolution requirements. We aim to present a comprehensive dataset suitable for physiological investigations of human sleep and benchmarking of algorithms with high spatiotemporal resolution requirements. Our dataset consists of 62 whole-night polysomnographic recordings of 40 subjects with no reported psychiatric or neurological conditions, obtained at the Neurophysiology Unit, University of Patras between 2007-2016. Acquired biosignals include EEG, electrooculogram, electrocardiogram and maseter electromyogram. A referential montage of 56 EEG channels allows for topographical analyses with fine spatial resolution. Moreover, a subset of 18 recordings is complemented by electrode positioning data captured using a and MRI scans, thus can be useful to EEG source localization. Original sampling rate of 2.5 KHz offers sub-millisecond temporal resolution, thus allows for challenging applications (e.g. EEG high frequency oscillations, heart rate variability analysis). The amplifier’s low frequency filter was 0.05 Hz, therefore data are suitable for investigations of broadband electrophysiological dynamics, including slow (<1 Hz) EEG
activity. Sleep scoring, artifact rejection and annotation of EEG microstructural events -namely, sleep spindles and K-complexes- have been performed by at least one and usually more experts. This dataset is intended to become available to the sleep research community on the basis of research
collaborations.