Photoplethysmographic-based automated sleep–wake classification using a support vector machine

Abstract

Objective: Sleep quality has a significant impact on human mental and physical health. The detection of sleep-wake states is thus of paramount importance in the study of sleep. The gold standard method for sleep-wake classification is multisensor based polysomnography (PSG) which is normally recorded in clinical setting. The main drawbacks of PSG are the inconvenience to the subjects, the impact of discomfort on normal sleep cycles, and its requirement for experts' interpretation. In contrast, we aim to design an automated approach for sleep-wake classification using wearable fingertip photoplethysmographic (PPG) signal. Approach: Time domain features are extracted from PPG and PPG based surrogate cardiac signal for sleepwake classification. The minimal redundancy maximal relevance feature selection algorithm is employed to reduce irrelevant and redundant features. Main results: Support vector machine (SVM) based supervised machine learning classifier is then used to classify sleep and wake states. The model is trained using 70% events (6575 sleep-wake events) from the dataset, and the rest 30% events (2818 sleep-wake events) are used for evaluating the performance of the model. Furthermore, the proposed model demonstrates a comparable performance (accuracy 81.10%, sensitivity 81.06%, specificity 82.50%, precision 99.37%, and F score 81.74%) with respect to the existing uni-modal and multi-modal methods for sleep-wake classification. Significance: This result advocates the potential of wearable PPG based sleep-wake classification. A wearable PPG based system would help in continuous, non-invasive monitoring of sleep quality.

Multi-stage sleep classification using photoplethysmographic sensor

“…SVM is a broadly used supervised machine-learning technique that points out the optimal separating hyperplane to distinguish the data by maximizing the margin between the classes in the feature space [ 52 ]. The polynomial kernel was selected for the SVM because it was shown to have the highest classification [ 25 ].…”

“…Clinically, PPG is a frequently used sensor to measure bloodoxygen saturation and heart rate. These physiological parameters are closely linked with sleep and hence have been proposed to identify two sleep stages: sleep and awake [21][22][23][24][25][26]. However, the accuracies reported in the above-mentioned studies suffer from biased sensitivity or specificity, which makes these unsuitable for computerized assessments.…”

Sleep classification using Consumer Sleep Technologies and AI: A review of the current landscape

“…Not every study specified all the details about the chosen device. For an example [28] only specified that they used a finger-worn PPG but specified sampling frequency. On the other end [23] used a Samsung Gear S2 smartwatch which contains multiple sensors but only used PPG which is why only PPG was reported as a sensors.…”

Photoplethysmography in Normal and Pathological Sleep

“…Motin et al [ 36 ] designed an automated approach, extracted from time domain features, for sleep–wake classification based on fingertip PPG signals (SpO2, surrogate cardiac signals). A support vector machine was used on a training dataset to teach the machine to distinguish sleep and wake stages, and this was then applied to 2818 sleep–wake events from PSG.…”