SleepEEGNet: Automated Sleep Stage Scoring with Sequence to Sequence Deep Learning Approach (1903.02108v1)

Abstract: Electroencephalogram (EEG) is a common base signal used to monitor brain activity and diagnose sleep disorders. Manual sleep stage scoring is a time-consuming task for sleep experts and is limited by inter-rater reliability. In this paper, we propose an automatic sleep stage annotation method called SleepEEGNet using a single-channel EEG signal. The SleepEEGNet is composed of deep convolutional neural networks (CNNs) to extract time-invariant features, frequency information, and a sequence to sequence model to capture the complex and long short-term context dependencies between sleep epochs and scores. In addition, to reduce the effect of the class imbalance problem presented in the available sleep datasets, we applied novel loss functions to have an equal misclassified error for each sleep stage while training the network. We evaluated the proposed method on different single-EEG channels (i.e., Fpz-Cz and Pz-Oz EEG channels) from the Physionet Sleep-EDF datasets published in 2013 and 2018. The evaluation results demonstrate that the proposed method achieved the best annotation performance compared to current literature, with an overall accuracy of 84.26%, a macro F1-score of 79.66% and Cohen's Kappa coefficient = 0.79. Our developed model is ready to test with more sleep EEG signals and aid the sleep specialists to arrive at an accurate diagnosis. The source code is available at https://github.com/SajadMo/SleepEEGNet.

Overview of SleepEEGNet: Automated Sleep Stage Scoring with Deep Learning

The paper "SleepEEGNet: Automated Sleep Stage Scoring with Sequence to Sequence Deep Learning Approach" introduces an innovative deep learning approach aimed at automating the sleep stage scoring process using electroencephalogram (EEG) signals. This is particularly important as manual scoring of sleep stages is both labor-intensive and susceptible to inter-rater variability. This paper describes a solution to this issue through SleepEEGNet, a novel deep learning architecture that employs convolutional neural networks (CNNs) and a sequence to sequence model to accurately classify sleep stages while addressing class imbalance issues inherently present in available datasets.

Methodological Insights

The SleepEEGNet model constructed in this paper leverages a combination of deep CNNs and bidirectional recurrent neural networks (BiRNNs) integrated within a sequence to sequence framework featuring an attention mechanism. CNNs act to extract time-invariant and frequency-specific features from EEG signals. The BiRNNs are employed to map complex temporal dependencies, simultaneously considering temporal data both in forward and backward directions, enhancing predictive accuracy by utilizing comprehensive contextual information from EEG sequences.

An essential innovation in this work is the introduction of new loss functions to mitigate the impact of class imbalance, a common challenge in sleep datasets. The mean false error (MFE) and mean squared false error (MSFE) loss functions ensure equal weighting of classification errors across class distributions. This strategic adaptation significantly enhances the model's robustness, particularly in scoring less frequent sleep stages such as N1.

Empirical Validation

The efficacy of the proposed SleepEEGNet model is demonstrated through robust empirical evaluation against sleep stage scoring on the Physionet Sleep-EDF datasets from 2013 and 2018. By achieving an overall accuracy of 84.26% on the 2013 dataset with an F1-score of 79.66% and a Cohen's kappa of 0.79, the model surpasses existing methods in the literature. Notably, this validation spans multiple single-channel EEG setups (Fpz-Cz and Pz-Oz), ensuring versatility and applicability across different recording conventions.

Application and Future Work

The practical implications of this research are significant, providing a viable tool for clinical settings to facilitate accurate, efficient, and reproducible sleep stage classification. This development holds promise for enhancing diagnostic processes in sleep medicine, potentially influencing the management of sleep-related disorders such as sleep apnea and insomnia.

The theoretical contributions of SleepEEGNet lie in its effective integration of deep learning paradigms with innovative loss function design, setting a precedent for addressing class imbalance in similar biomedical applications. Future avenues for this research could explore multimodal integration with other physiological signals such as EOG and EMG, aiming for even higher accuracy and broader applicability. Additionally, further exploration into the temporal evolution of sleep stages could refine model interpretation and enhance clinical insights derived from automated scoring algorithms.

In conclusion, SleepEEGNet represents a substantial contribution to the domain of neural signal processing and computational sleep medicine, offering a sophisticated, data-driven approach to a traditionally manual process.