On-Device Integrated PPG Quality Assessment and Sensor Disconnection/Saturation Detection System for IoT Health Monitoring

• Published in: IEEE transactions on instrumentation and measurement 2020-09, Vol.69 (9), p.6351-6361
• Main Authors: Narendra Kumar Reddy, Gangireddy, Sabarimalai Manikandan, M., Narasimha Murty, N. V. L.
• Format: Article
• Language: English
• Subjects: Detectors; Entropy (Information theory); False alarm reduction; False alarms; Feature extraction; Flash memory (computers); Indexes; Internet of Things; Internet-of-Things (IoT) medical devices and smartphone health monitoring; Kurtosis; Microprocessors; Monitoring; motion artifact detector; Noise measurement; photoplethysmography (PPG); PPG signal quality assessment (SQA); Quality assessment; Random noise; Saturation; Segments; Sensors; Signal quality
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2020.2971132

In this article, we propose a new on-device photoplethysmography (PPG) signal quality assessment (SQA) system based on the first-order (FO) predictor coefficient (PC) of differenced sensor (DS) signal that is capable of significantly reducing false alarms and timely notifying sensor disconnection and saturation in unsupervised health monitoring. The proposed system consists of the DS signal computation, random noise addition, PC estimation, and hierarchical decision rules. The proposed system is evaluated by using varieties of sensor signals taken from standard databases and the PPG signals recorded using three PPG sensing modules available at our Biomedical System Laboratory. Results show that the proposed method achieves correct detection rates of 98.22%, 80.09%, and 99.79% for noise-free PPG segments, motion artifacts corrupted segments, and pulse-free segments, respectively, and also outperforms other SQA methods. The proposed method had a false alarm reduction rate of 90.71% and a missed acceptable quality rate of 1.78%. Real-time implementation on the Arduino Due computing platform with a 32-bit Cortex-M3 processor, 512-kB flash memory, and 96-kB data memory demonstrates that the proposed on-device SQA system consumes 0.141 mJ with an overall accuracy of 93.21%, whereas the kurtosis and Shannon entropy (KSE) and the fiducial and template (FT)-based methods consume 109.08 and 6.71 mJ with overall accuracies of 39.24% and 74.13%, respectively. The proposed system has great potential for developing energy-efficient accurate and reliable Internet of Things (IoT) and smartphone-based health monitoring systems.