On-Device Signal Quality Guided and Embedded Physiologic Information for High Fidelity Continuous PPG Compression

Photoplethysmography (PPG) is widely popularized in wearable gadgets for continuous health monitoring. The main challenge before the PPG-based measurements is motion artifact (MA) corruption, which often suppress vital physiological information like heart rate (HR) and respiratory rate (RR), which makes their extraction difficult and sometime erroneous. This work introduces a new approach of PPG multiclass signal quality assessment (SQA) using reservoir computing (RC) and tunable ${Q}$ -factor wavelet transform (TQWT) for PPG compression. To ensure the post compression clinical usage, HR and RR are embedded in the compressed bit stream, at every 10 s PPG frame. For the evaluation, a total of 200 records were utilized from three public datasets, namely, MIMIC II/III waveform, PPG RR benchmark (PRRB), BIDMC, and 20 volunteers at our laboratory, spanning a total duration of 7740 min. The technique showed good on-device (ARM v6 controller) performance on SQA (accuracy of 99.97%) and PPG compression (CR of 38.28 and percentage root mean squared difference (PRD) of 3.04 at 125 Hz sampling) using MIMIC data, which are competitive with published works. The low end-to-end latency (51 ms) and run time memory (19.99 kB) per 1 s PPG shows the usefulness for real-time deployment. It was also showed that embedding the physiologic information in compressed data may reduce the post compression processing error in RR and HR by an amount of 5.61% and 31.23% respectively in corrupted PPG. The proposed technique can be utilized in high fidelity PPG monitoring applications under ambulatory measurement.