Automatic sleep stage classification based on subcutaneous EEG in patients with epilepsy

Abstract

Background: The interplay between sleep structure and seizure probability has previously been studied using electroencephalography (EEG). Combining sleep assessment and detection of epileptic activity in ultralong-term EEG could potentially optimize seizure treatment and sleep quality of patients with epilepsy. However, the current gold standard polysomnography (PSG) limits sleep recording to a few nights. A novel subcutaneous device was developed to record ultralong-term EEG, and has been shown to measure events of clinical relevance for patients with epilepsy. We investigated whether subcutaneous EEG recordings can also be used to automatically assess the sleep architecture of epilepsy patients.

Method: Four adult inpatients with probable or definite temporal lobe epilepsy were monitored simultaneously with long-term video scalp EEG (LTV EEG) and subcutaneous EEG. In total, 11 nights with concurrent recordings were obtained. The sleep EEG in the two modalities was scored independently by a trained expert according to the American Academy of Sleep Medicine (AASM) rules. By using the sleep stage labels from the LTV EEG as ground truth, an automatic sleep stage classifier based on 30 descriptive features computed from the subcutaneous EEG was trained and tested.

Results: An average Cohen's kappa of [Formula: see text] was achieved using patient specific leave-one-night-out cross validation. When merging all sleep stages into a single class and thereby evaluating an awake-sleep classifier, we achieved a sensitivity of 94.8% and a specificity of 96.6%. Compared to manually labeled video-EEG, the model underestimated total sleep time and sleep efficiency by 8.6 and 1.8 min, respectively, and overestimated wakefulness after sleep onset by 13.6 min.

Conclusion: This proof-of-concept study shows that it is possible to automatically sleep score patients with epilepsy based on two-channel subcutaneous EEG. The results are comparable with the methods currently used in clinical practice. In contrast to comparable studies with wearable EEG devices, several nights were recorded per patient, allowing for the training of patient specific algorithms that can account for the individual brain dynamics of each patient. Clinical trial registered at ClinicalTrial.gov on 19 October 2016 (ID:NCT02946151).

Keywords: Automatic sleep scoring; Epilepsy; Sleep; Subcutaneous EEG; Wearable EEG.

Conflict of interest statement

SWG, JDH and MCH are employed by UNEEGTM medical A/S. MHL and SW is partially funded by UNEEGTM medical A/S. TWK consults for UNEEGTM medical A/S. The remaining authors declare no competing interests.

Figures

Fig. 1

Box plot of Cohen’s kappa values. The midline in the boxes represent the medians, and the dots represent the means. Red crosses are outliers. The mean value ± standard deviation of the mean for the five-class problem: κPS.=0.78±0.02, κLONO=0.74±0.02 and κexpert=0.66±0.04. Mean value ± standard deviation of the mean for the two-class problem: κPS.=0.85±0.03, κLONO=0.82±0.03 and κexpert=0.81±0.04. The horizontal lines represent intervals of the level of agreement as interpreted by McHugh et al. [18]

Fig. 2

Representative night: the second night of patient B. Top panel: spectrogram of the proximal subcutaneous EEG channel (P–C). Middle panel: spectrogram of the corresponding scalp channel (P7–T7). Bottom panel: manually scored hypnogram based on scalp EEG and the predicted hypnograms by the PS and LONO algorithm and the human expert

Fig. 3

Confusion matrices for the five- and two-class problems. Each entry in the matrices provides the percentage P of epochs known to belong to class i that were classified as belonging to class j, for i,j∈{1,…NumberOfClasses, and the raw count. The percentage P in the diagonal equals the class sensitivity. The coloring reflects the magnitude of P, which ranges from 0 to 100 %

Fig. 4

Comparison of ground truth sleep measures to estimated sleep measures. The blue squares indicate results from the PS algorithm, the red circles are the LONO algorithm and the yellow diamonds are the human expert. Left: scatter plot with Deming regression line, slope of regression line (β) and Pearson’s correlation coefficient (r). Right: Bland–Altman plots. The solid line is the mean difference, and the dotted lines are 1.96 times the standard deviation of the mean

Fig. 5

Illustration of the subcutaneous recordings system. Left: illustration of the implant and the beta-version of the external device used to collect data in the present study. The placement of the Proximal (P), Center (C) and Distal (D) electrodes are indicated by the letters. The length of the implant is approximately 11 cm. Right: illustration of the commercially available device. The device is worn under the shirt and secured in place by a magnet (gray circle)

Fig. 6

Manually scored hypnograms based on scalp EEG. The five tick marks on the y-axis represent (from top to bottom) wake, REM sleep, N1, N2 and N3. REM sleep is marked with a red, bold line. Three nights were recorded for each of patients A, C and D, and two nights were recorded for patient B