Lightweight Time-Domain HRV Index Extraction from Compressive ECG Measurements for Resource Constrained Edge Health Computing Devices

Electrocardiogram (ECG) derived heart rate variability (HRV) analysis is widely used in both physiological and psychological monitoring. For HRV analysis, an automatic R-peak to R-peak interval (RRI) determination is most essential and that is performed by processing the ECG signal. Continuous HRV analysis is most essential for predicting health problems and stress levels, but HRV analysis devices are restricted with limited battery capacity. To increase the battery lifetime of monitoring devices, we present a simple energy-efficient R-peak detection approach by directly processing a few compressive measurements instead of processing the ECG signal sampled at the Nyquist rate. The proposed CS-ECG based HRV analysis method consists of: compressed sensing (CS) based data reduction, straightforward RRI determination without the use of searchback algorithm with sets of thresholds, and the HRV parameter estimation. The accuracy and robustness of CS-ECG based time-domain HRV index extraction (7 HRV indexes) is evaluated using the MIT-BIH arrhythmia database, MIT-BIH polysomnographic database (slpdb) and Apnea-ECG database. Evaluation results show that the estimated HRV parameters from RR intervals in the CS domain are comparable with the HRV parameters estimated from the RR intervals extracted from the uncompressed ECG signal. Using deterministic binary block diagonal (DBBD) based measurement generation with a compression factor of 4, the CS-ECG based RRI determination can reduce computational resources by 75% without significant estimation error. Thus, the CS-ECG based HRV analysis method has potential to reduce overall energy consumption, memory space, and processing time by a factor of 4 as compared to the HRV analysis method with an uncompressed ECG signal.