Sleep Stage Task

Work with saved features

For a saved-feature workflow, start with the staging baseline, which describes input records, preprocessing and checkpoint evaluation. Continue to staging training to train a model with an explicit split and configuration.

Overview

The objective of sleep stage assessment is to determine the different stages of sleep (N1, N2, N3, REM, and wake) over the course of a night. The de facto standard for sleep stage assessment is polysomnography (PSG) that requires collecting numerous multi-site physiological signals and a trained sleep technician to perform the assessment. The technician uses the physiological signals to determine the sleep stage based on the American Academy of Sleep Medicine (AASM) guidelines. The identification of stages is possible due to specific electrophysiological signatures present and recorded by EEG, EOG, and EMG signals.

The focus of this project is to develop a model that can perform sleep stage assessment using only a small subset of the signals that can be captured from same body location (e.g. wrist) that is viable for long-term use. In particular, the current models are trained on physiological signals that can be captured from the wrist for fitness band and smartwatch applications. Traditionally, wrist-based sleep classification has been performed using actigraphy, which is a method of monitoring gross motor activity using an accelerometer. However, actigraphy is not a reliable method for sleep stage assessment. The current model is designed to perform sleep stage assessment using a combination of motor activity, cardiovascular signals (e.g. heart rate), and derived resipiratory signals (e.g. respiration rate).

Sleep Stages

Sleep stages consists of wake (W), non-rapid eye movement (NREM) sleep, and rapid eye movement (REM) sleep. NREM sleep is further divided into three stages: N1, N2, and N3. N1 and N2 are considered light sleep, while N3 is considered deep sleep. REM sleep is considered the dreaming stage of sleep. During the course of a night, a subject will undergo multiple rounds of sleep cycles, each composed of the individual stages. The duration of each cycle and length of each stage various throughout the night. Each sleep cycle typically lasts around 90 to 120 minutes.

Stage	Share
N1	5%
N2	55%
N3	10%
REM	30%

Example sleep-stage proportions. The distribution varies between nights and individuals.

This is the time before or after bed when individual is awake.

Key Characteristics

Alpha waves on the EEG
No K-complexes or sleep spindles
Eye movements are normal
Respiratory rate and heart rate are normal
Body movements are common

Model Zoo

The following table provides the latest performance and accuracy results for pre-trained models. Additional result details can be found in Model Zoo → Stage.

NAME	LOCATION	# CLASSES	MODEL	PARAMS	FLOPS	ACCURACY	AP
SS-2-TCN-SM	Wrist	2	TCN	10K	1.7M/hr	88.9%	96.0%
SS-3-TCN-SM	Wrist	3	TCN	11K	1.9M/hr	84.4%	91.8%
SS-4-TCN-SM	Wrist	4	TCN	26K	4.2M/hr	75.5%	81.6%
SS-5-TCN-SM	Wrist	5	TCN	28K	4.6M/hr	68.4%	74.1%

Target Classes

Below outlines the classes available for sleep stage classification. When training a model, the number of classes, mapping, and names must be provided.

CLASS	LABELS
0	WAKE
1	STAGE1 (N1)
2	STAGE2 (N2)
3	STAGE3 (N3)
4	STAGE4 (N4)
5	REM
6	NOISE

Class Mapping

Below is an example of a class mapping for a 4-class sleep stage model. The class map keys are the original class labels and the values are the new class labels. Any class not included will be skipped.

{
    "num_classes": 4,
    "class_names": ["AWAKE", "LIGHT", "DEEP", "REM"],
    "class_map": {
        "0": 0,
        "1": 1,
        "2": 1,
        "3": 2,
        "4": 2,
        "5": 3
    }
}

Limitations

Inferring sleep stage without electrophysiological signals is an extremely challenging task. Especially when considering a highly noisy ambulatory environment. Furthermore, the limited time spent in certain sleep stages such as N1 makes it difficult to accurately classify these stages. This is highlighted by the limited interrater reliability of sleep stage scoring. A recent paper by Kapur et al., 2021 performed a meta-analysis of 101 studies and found that the interrater reliability for stage N1 sleep was only fair, while the interrater reliabilities for stage N2 and N3 sleep were moderate. The paper found the Cohen’s kappa for manual, overall sleep scoring was 0.76, indicating substantial agreement (95% confidence interval, 0.71–0.81; P < .001). By sleep stage, the figures were 0.70, 0.24, 0.57, 0.57, and 0.69 for the W, N1, N2, N3, and R stages, respectively. The interrater reliabilities for stage N2 and N3 sleep were moderate, and that for stage N1 sleep was only fair. Taking this into account, we should not expect to achieve perfect sleep stage classification.