Robust Estimation of Pulse Rate from a Wrist-type PPG During Intensive Exercise.

Estimation of pulse rate from a wrist-type PPG during motion is a notoriously difficult problem because of the presence of motion artifact (MA) which corrupts the signal in both the time and frequency domains. In this paper, we propose a new method for deriving pulse rate under intense exercise conditions which employs Ensemble Empirical Mode Decomposition and power spectral analysis to extract the pulsatile component of the signal. The method was validated on an openly available database containing PPG and ground-truth ECG-derived pulse rate measurements from 12 subjects during a running experiment. Our proposed technique showed a high estimation accuracy with a mean absolute error of 2.14 bpm over the entire database and a correlation coefficient between the estimates and the ground truth of 0.98. Our approach matched the performance of the state-of-the-art TROIKA framework without utilizing simultaneously recorded accelerometry data to remove the MA component. With over 97.5% of estimates within a 10% margin from the ground truth, our technique shows a lot of potential for inclusion in next generation wrist-worn wearable monitors in both sports and clinical settings.