A Novel Deep Learning Model Using an LSTM Algorithm to Automate Sleep Staging from Sleep EEG Data.

A majority of Americans suffer from sleep deprivation. Nearly thirty percent of adults get under six hours of sleep, and only thirty percent of high schoolers get the recommended eight hours of sleep daily.1 The prevalence of sleep-related health issues combined with the problem of sleep deprivation makes it crucial to discover how a person sleeps. The data that a Polysomnography (PSG - the standard measurement for sleep quality) provides, such as electroencephalogram (EEG) signals, are used to diagnose sleep disorders. In this paper, we delve into the separate components of sleep EEG and discuss a novel model that automates sleep staging. We trained and tested our model using a PhysioNet dataset containing one hundred ninety-seven wholenight polysomnographic sleep recordings. The architecture of our model is based on a Long Short-Term Memory (LSTM) model, a type of Recurrent Neural Network commonly used in sequential data analysis and time series classifications. Though there are some deficiencies in our model, overall, we had promising results, with a mean F1 score of 0.79 and a mean accuracy of 88%, with the highest accuracy being 98%. Our findings showed an upward trend in accuracy for the N2, N3, and Wake stages and a decrease in accuracy for the Wake and N1 Sleep stages. This paper examines experimental results, interpretations, potential improvements, and future modifications to improve the results from our current model. The need for highly accurate sleep staging models is crucial, as it can prevent a potential sleep disorder epidemic and advance the diagnosis of fundamental anomalies in sleep EEG. [ABSTRACT FROM AUTHOR]