Measuring sleep quality from heart and muscle signals
- Article
- Health
Most of what we know about a person’s sleep comes from asking them. Questionnaires such as the Pittsburgh Sleep Quality Index are cheap and widely used, but they depend on memory and self-judgment. A 2023 paper by Nada Fitrieyatul Hikmah, Rachmad Setiawan, and Mohammad Daffa Gunawan of the Department of Biomedical Engineering at Institut Teknologi Sepuluh Nopember in Surabaya, Indonesia, asks whether body signals recorded after waking can do the same job more objectively.
What they measured
The study worked from the idea that poor sleep shows up as fatigue, and fatigue leaves traces in the heart and the muscles. The team recruited ten male subjects aged 21 to 22 with no chronic illness. For each subject they recorded two signals, five times on different occasions:
- An electrocardiogram (ECG), five minutes long, from electrodes on the arms and left leg. From this they computed heart rate variability, the small fluctuations in the time between beats, which reflects the balance of the autonomic nervous system.
- An electromyogram (EMG), ten seconds long, from surface electrodes on the neck extensor, trapezius, and deltoid muscles. From this they computed the mean power frequency, which falls as a muscle tires.
After each recording the subject filled in the Fatigue Severity Scale questionnaire. That score became the label the models had to predict.
The features and the models
The authors say earlier work used only a few standard heart rate variability measures. They extracted a fuller set: time-domain values (SDNN, RMSSD, SDSD, pNN50), the low-to-high frequency power ratio from a spectral analysis, and two nonlinear measures (SD1 and SD2) from a Poincaré plot, plus the EMG mean power frequency and body mass index. These fed four supervised classifiers: logistic regression, random forest, k-nearest neighbors, and a support vector machine.
Results
Logistic regression, random forest, and k-nearest neighbors each reached 93 percent accuracy. The plain support vector machine did poorly at 40 percent, but after tuning its parameters with a grid search it classified every test case correctly.
Caveats
Ten subjects and fifty recordings is a small dataset, and all the subjects were young men, so the perfect score should be read as a proof of concept rather than a general result. The value of the paper is in the pipeline: a short, noninvasive recording session, a fixed set of features, and standard classifiers that together track a questionnaire score. With a larger and more varied sample, the same setup could give clinicians a quick objective check on sleep quality.