JMIR Publications

ABSTRACT

Background:

With the development and promotion of wearable devices and their mobile health (mHealth) applications, physiological signals have become a research hotspot. However, noise is complex in signals during daily lives, making it hard to analyze the signals automatically and resulting in a high false alarm rate. How to screen out the high-quality segments of the signals from huge data volumes remains a problem. Signal quality assessment (SQA) can help to advance the valuable information mining of signals as well as the mHealth applications.

Objective:

The first aim of this study was to establish an SQA algorithm based on the unsupervised isolation forest model and a small amount of labeled data to classify the Electrocardiogram (ECG) and respiration signal quality into 3 different levels. A second aim was to apply the SQA algorithm to real-world data to demonstrate that the algorithm has the potential to reduce the false alarms caused by poor signal quality.

Methods:

Data used in this study was measured by a wearable device, SensEcho, from healthy subjects and patients in general wards. The observation windows for ECG and respiratory signals were 10s and 30s respectively. The experimental procedure consisted of 4 key steps: the unlabeled training set was used to train the models. The validation set and test set were labeled according to the evaluation criteria that had been set based on clinical experience and characteristics of signals. The validation set consisted of 3460 & 2086 windows of ECG and respiratory signals respectively while the test set was made up of 4686 & 3341. The validation set was used to determine the best classification performance threshold values while the test set was used to test the generalization performance of the threshold values. The algorithm was compared with classic supervised models and the case validation was conducted to observe the results of the algorithm qualitatively. Then, the algorithm was applied to 1144 cases of ECG signal collected from patients in the general wards of the Hyperbaric Oxygen Department and the defined false alarm proportion was calculated.

Results:

The quantitative results showed that the ECG SQA model reached 94.97% & 95.58% accuracy on the validation set and test set respectively, while respiratory SQA model reached 81.06% & 86.20% accuracy on the validation set and test set respectively. These results were of the same level as the results of supervised models. The example case showed that the algorithm was able to mark the poor-quality segments of signals correctly, demonstrating that the algorithm had a good generalization capacity. The expected algorithm application results were obtained, which indicated that the median [Q1,Q3] for good, acceptable and unacceptable quality proportion was 90.0% [81.4%,95.9%], 4.8% [2.1%,8.0%], 4.0% [1.1%,9.3%] respectively. The detection accuracy of some type of the arrhythmia is likely to be influenced by signal quality. The arrhythmia false alarms caused by poor signal quality can be reduced significantly with the help of the SQA algorithm.

Conclusions:

This study verified the feasibility of applying the anomaly detection unsupervised model to SQA. The performance of the algorithm was good. The algorithm was very close to practical use, and its applications include reducing the false alarm rate of the device and selecting signal segments that can be used for further research.