Assessing Electrocardiogram and Respiratory Signal Quality of a Wearable Device (SensEcho): Semisupervised Machine Learning-Based Validation Study

• DOI: 10.2196/25415
• Publication Date: 2021-08-12T14:31:07Z
• Authors: Haoran Xu; Wei Yan; Ke Lan; Chenbin Ma; Di Wu; Anshuo Wu; Zhicheng Yang; Jiachen Wang; Yaning Zang; Muyang Yan; Zhengbo Zhang
• Keywords: Wearable Technology

Background With the development and promotion of wearable devices their mobile health (mHealth) apps, physiological signals have become a research hotspot. However, noise is complex in obtained from daily lives, making it difficult to analyze automatically resulting high false alarm rate. At present, screening out high-quality segments huge-volume data with few labels remains problem. Signal quality assessment (SQA) essential able advance valuable information mining signals. Objective The aims this study were design an SQA algorithm based on unsupervised isolation forest model classify signal into 3 grades: good, acceptable, unacceptable; validate labeled sets; apply real-world evaluate its efficacy. Methods Data used collected by device (SensEcho) healthy individuals patients. observation windows for electrocardiogram (ECG) respiratory 10 30 seconds, respectively. In experimental procedure, unlabeled training set was train models. validation test sets according preset criteria classification performance quantitatively. consisted 3460 2086 ECG signals, respectively, whereas made up 4686 3341 also compared self-organizing maps (SOMs) 4 classic supervised models (logistic regression, random forest, support vector machine, extreme gradient boosting). One case illustrated show application effect. then applied 1144 cases patients detected arrhythmia alarms calculated. Results quantitative results showed that achieved 94.97% 95.58% accuracy sets, 81.06% 86.20% superior SOM moderate when example correctly even there pathological changes indicated some specific types such as tachycardia, atrial premature beat, ventricular beat could be significantly reduced help algorithm. Conclusions This verified feasibility applying anomaly detection SQA. scenarios include reducing rate selecting can further research.