Coon, W. G.; Zerr, P.; Milsap, G.; Sikder, N.; Smith, M.; Dresler, M.; Reid, M.

The increasing availability of wearable forehead EEG devices, such as the Hypnodyne ZMax, open-source DCM, CGX PatchEEG, and many others, has significantly expanded opportunities for convenient, at-home sleep monitoring. However, most publicly available sleep classifiers are trained on scalp EEG from traditional polysomnography (PSG) data, and thus generalize poorly to forehead EEG due to differences in electrode placement, referencing montages, and higher susceptibility to artifacts from user movements and electrode displacement. Conventional classifiers typically do not explicitly account for these artifacts, resulting in inaccurate and misleading sleep stage scoring. To address this gap, we developed a suite of artifact-aware sleep stage classifiers trained specifically using forehead EEG data, leveraging two comprehensive datasets --Wearanize+ and Donders2022-- that contain concurrent forehead EEG and clinical PSG recordings. We further introduce two classifier variants: one optimized for real-time applications that operates directly on raw EEG amplitudes, and another optimized for offline analysis utilizing normalized EEG signals. Validation results indicate robust and reliable classification performance across standard sleep stages (Wake, N1, N2, N3, REM), along with effective identification of artifact epochs. Importantly, the developed classifiers generalize well to forehead EEG devices beyond the original training platform. These validated classifiers are freely available to the sleep research community through the open-source ezscore-f package, providing versatile and practical tools for forehead EEG-based sleep stage analysis.