Expert-level sleep scoring with deep neural networks

Biswal, Siddharth; Sun, Haoqi; Goparaju, Balaji; Westover, M. Brandon; Sun, Jimeng; Bianchi, Matt T.

doi:10.1093/jamia/ocy131

Cited by 200 publications

(144 citation statements)

References 37 publications

Supporting

Mentioning

130

Contrasting

Unclassified

Order By: Relevance

“…In contrast to previous studies, the 2018 PhysioNet Challenge offered us a unique opportunity to truly evaluate the performances and compare cutting-edge methods on a large external hidden test set of 989 samples 23 . In addition, we demonstrate that deep convolutional neural networks trained on full-length records and multiple physiological channels have the best performance in detecting sleep arousals, which are quite different from current approaches extracting features from short 30-second epochs 25,27,30 . Beyond sleep arousals, we propose that the U-Net architecture used in DeepSleep can be adapted to other segmentation tasks such as sleep staging.…”

Section: Discussionmentioning

confidence: 57%

“…S2G-H ). It has also been reported that neural network approaches significantly outperformed classical machine learning methods, including random forest, logistic regression 25 , support vector machine, and linear models 26 . In fact, 8 out of the top 10 teams used neural network models in the 2018 PhysioNet Challenge (red blocks in Fig.…”

Section: Resultsmentioning

confidence: 99%

“…We also compared DeepSleep with recent state-of-the-art methods in sleep stage scoring. These methods extracted features from 30-second epochs through short-time Fourier transform (STFT) 27,28 or Thomson’s multitaper 25,29 . They were originally designed for automatic sleep staging and we applied them to the task of detecting sleep arousals on the same 2018 PhysioNet data.…”

Section: Resultsmentioning

confidence: 99%

See 2 more Smart Citations

DeepSleep: Fast and Accurate Delineation of Sleep Arousals at Millisecond Resolution by Deep Learning

Guan

2019

Preprint

View full text Add to dashboard Cite

Sleep arousals are transient periods of wakefulness punctuated into sleep. Excessive sleep arousals are associated with many negative effects including daytime sleepiness and sleep disorders. High-quality annotation of polysomnographic recordings is crucial for the diagnosis of sleep arousal disorders. Currently, sleep arousals are mainly annotated by human experts through looking at millions of data points manually, which requires considerable time and effort. Here we present a deep learning approach, DeepSleep, which ranked first in the 2018 PhysioNet Challenge for automatically segmenting sleep arousal regions based on polysomnographic recordings. DeepSleep features accurate (area under receiver operating characteristic curve of 0.93), high-resolution (5-millisecond resolution), and fast (10 seconds per sleep record) delineation of sleep arousals.

show abstract

Section: Discussionmentioning

confidence: 57%

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

DeepSleep: Fast and Accurate Delineation of Sleep Arousals at Millisecond Resolution by Deep Learning

Guan

2019

Preprint

View full text Add to dashboard Cite

show abstract

“…Deep learning approaches can be used to accurately estimate sleep states. We previously showed that deep neural networks can learn to score conventional sleep stages based on EEG signals obtained during overnight PSG with an accuracy of 87.5% and a Cohen's kappa of 0.805, comparable to the performance of human sleep scoring experts 17 . Here we develop deep neural networks using ECG and/or respiratory signals to classify sleep stages.…”

Section: Introductionmentioning

confidence: 85%

Sleep staging from electrocardiography and respiration with deep learning

Sun

Ganglberger

Panneerselvam

et al. 2019

Sleep

Self Cite

View full text Add to dashboard Cite

Study Objective:Sleep is reflected not only in the electroencephalogram but also in heart rhythms and breathing patterns. Therefore, we hypothesize that it is possible to accurately stage sleep based on the electrocardiogram (ECG) and respiratory signals.Methods: Using a dataset including 8,682 polysomnographs, we develop deep neural networks to stage sleep from ECG and respiratory signals. Five deep neural networks consisting of convolutional networks and long short-term memory networks are trained to stage sleep using heart and breathing, including the timing of R peaks from ECG, abdominal and chest respiratory effort, and the combinations of these signals.Results: ECG in combination with the abdominal respiratory effort achieve the best performance for staging all five sleep stages with a Cohen's kappa of 0.600 (95% confidence interval 0.599 -0.602); and 0.762 (0.760 -0.763) for discriminating awake vs. rapid eye movement vs. non-rapid eye movement sleep. The performance is better for young participants and for those with a low apnea-hypopnea index, while it is robust for commonly used outpatient medications. Conclusions:Our results validate that ECG and respiratory effort provide substantial information about sleep stages in a large population. It opens new possibilities in sleep research and applications where electroencephalography is not readily available or may be infeasible, such as in critically ill patients. Deep Network ArchitectureWe trained five deep neural networks based on the following input signals and their combinations: 1) ECG; 2) CHEST (chest respiratory effort); 3) ABD (abdominal respiratory effort); 4) ECG+CHEST; and 5) ECG+ABD. Each deep neural network contained a feed-forward convolutional neural network (CNN) which learned features pertaining to each epoch, and a recurrent neural network (RNN), in this case long-short term memory (LSTM), to learn temporal patterns among consecutive epochs.The CNN of the network is similar to that in Hannun et al. 20 . As shown in Figure 1A and Figure 1B, the network for a single type of input signal, i.e. ECG, CHEST or ABD, consists of a convolutional layer, several residual blocks and a final output block. For a network with both ECG and CHEST/ABD as input signals ( Figure 1C), we first fixed the weights of the layers up to the 9 th residual block (gray) for the ECG network and similarly fixed up to the 5 th residual block (gray) for the CHEST/ABD network, concatenated the outputs, and then fed this concatenation into a subnetwork containing five residual blocks and a final output block. The numbers of fixed layers were chosen so that the outputs of layers from different modalities have the same shape (after padding zeros), and were then concatenated.The LSTM of the network has the same structure for different input signals. It is a bi-directional LSTM, where the context cells from the forward and backward directions are concatenated. For the network

show abstract

“…Uses of convolutional and recurrent neural networks have been shown to achieve state-of-the-art performance in various fields [17,32], including sleep analysis [3,33,34]. A convolutional neural network (CNN) works by taking a static input such as a signal or image, and the CNN processes it with a network of filters.…”

Section: Classification With Convolutional and Lstm Neural Networkmentioning

confidence: 99%

Automatic detection of cortical arousals in sleep and their contribution to daytime sleepiness

Brink-Kjær

Olesen

Peppard

et al. 2020

Clinical Neurophysiology

View full text Add to dashboard Cite

Cortical arousals are transient events of disturbed sleep that occur spontaneously or in response to stimuli such as apneic events. The gold standard for arousal detection in human polysomnographic recordings (PSGs) is manual annotation by expert human scorers, a method with significant interscorer variability. In this study, we developed an automated method, the Multimodal Arousal Detector (MAD), to detect arousals using deep learning methods. The MAD was trained on 2,889 PSGs to detect both cortical arousals and wakefulness in 1 second intervals. Furthermore, the relationship between MAD-predicted labels on PSGs and next day mean sleep latency (MSL) on a multiple sleep latency test (MSLT), a reflection of daytime sleepiness, was analyzed in 1447 MSLT instances in 873 subjects. In a dataset of 1,026 PSGs, the MAD achieved a F1 score of 0.76 for arousal detection, while wakefulness was predicted with an accuracy of 0.95. In 60 PSGs scored by multiple human expert technicians, the MAD significantly outperformed the average human scorer for arousal detection with a difference in F1 score of 0.09. After controlling for other known covariates, a doubling of the arousal index was associated with an average decrease in MSL of 40 seconds (β = -0.67, p = 0.0075). The MAD outperformed the average human expert and the MAD-predicted arousals were shown to be significant predictors of MSL, which demonstrate clinical validity the MAD.

show abstract

Expert-level sleep scoring with deep neural networks

Cited by 200 publications

References 37 publications

DeepSleep: Fast and Accurate Delineation of Sleep Arousals at Millisecond Resolution by Deep Learning

DeepSleep: Fast and Accurate Delineation of Sleep Arousals at Millisecond Resolution by Deep Learning

Sleep staging from electrocardiography and respiration with deep learning

Automatic detection of cortical arousals in sleep and their contribution to daytime sleepiness

Contact Info

Product

Resources

About