0085 SleepInceptionNet: A Deep Learning Algorithm for Real-Time Sleep Stages Scoring Using Single-channel EEG

Abstract Introduction Most of the current automatic polysomnography sleep staging methods use multi-signal data and require a sequence preceding following epochs to score stage specific epoch, which may not be desirable for analysis in real-time and/or free-living conditions. We developed deep learning-based algorithm, namely SleepInceptionNet, that is designed each epoch using only single-channel electroencephalogram (EEG) within epoch. Methods Polysomnography 883 participants (937,975 thirty-second epochs) Multi-Ethnic Study Atherosclerosis (MESA) obtained from National Sleep Research Resource (NSRR) were randomly divided into separate training/validation set 194 test 689 participants. Each 30-second raw central EEG channel signal was transformed time-frequency domain images continuous wavelet transform method. InceptionV3 convolutional neural network structure trained tuned on training classify one five stages Wake, N1, N2, N3, or rapid eye movement (REM) sleep. Results Compared ground truth manually scored stages, SleepInceptionNet achieved an overall kappa agreement 0.690 weighted accuracy 0.897. The model showed (mean±SD across participants) 0.940±0.067 detecting 0.883±0.047 0.845±0.055 0.939±0.038 0.930±0.038 REM epochs, reference polysomnography. Conclusion high with epoch-by-epoch classification stages. This study demonstrates viability real-time, accurate EEG, could have variety applications such as delivery on-demand interventions during Support (If Any) MESA Ancillary funded by NIH-NHLBI Association Disorders Cardiovascular Health Across Ethnic Groups (RO1 HL098433). supported NHLBI (HHSN268201500003I, N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01- HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169) NCATS (UL1-TR-000040, UL1-TR-001079, UL1-TR-001420). NSRR (R24 HL114473, 75N92019R002).