Influence of Photoplethysmogram Signal Quality on Pulse Arrival Time during Polysomnography

Figure 1. Examples of MESA multichannel physiological signals recorded during (a) OSA and (b) hypopnea episodes: electroencephalogram (EEG), electrocardiogram (ECG), photoplethysmogram (PPG), arterial blood oxygen saturation (SpO2), nasal cannula flow, abdomen, and thorax movements.

Figure 2. The block diagram of the PPG signal-quality assessment algorithm for robust PAT estimation.

Figure 3. A low-complexity detection of T waves from the ECG signal by using additional filtering.

Figure 4. The segmentation of low-quality PPG pulse waves: (a) ECG signal with detected T waves; (b) PPG signal with detected pulse onsets and estimated maximum values of the PPG signal envelope, ePPG (black circles); (c) the second derivative of the PPG signal envelope after offsetting, |e″PPG|o (shown in green) was calculated from the PPG signal envelope, ePPG (black circles); (d) the identification of low-quality pulses according to thresholds θ1 and θ2 (red stars indicate low- quality pulses).

Figure 5. The relationship between PPG signal quality and PAT: (a) ECG signal; (b) PPG signal with labeled signal quality (SQI); (c) PAT variations estimated as time intervals between ECG R peaks and PPG systolic peaks—PAT before post-processing, PATb, and PAT after post-processing, PATa.

Figure 6. An example of PAT variations during OSA episodes: (a) respiration flow signal with labeled OSA episodes; (b) SpO2 variations; (c) PAT before post-processing, PATb; (d) PAT after post-processing, PATa.

Figure 7. Examples of PPG signals with labeled poor-quality segments obtained by ECG-guided and TM-based algorithms: (a) wake, (b) oxygen desaturation, and (c) arousal.

Figure 8. The proportion of PPG pulse waves classified as high quality by the two signal-quality assessment algorithms with estimated matched-pairs rc values: ECG-guided and TM-based approaches. Results are shown (a): whilst awake and for different sleep stages; (b) for different events related to sleep-disordered breathing. p < 0.001 and rc < 0.30 is marked *, p < 0.001 and rc≥ 0.30—**, p < 0.001 and rc≥ 0.50—***.

Figure 9. An example of PAT variations during regular and irregular rhythms: (a) PPG signal; (b) detected episodes of irregular rhythm (1—irregular rhythm, 0—regular rhythm); (c) PPG signal quality estimated by the ECG-guided algorithm, SQI (SQI = 1—high-quality, SQI = 0—low-quality); (d) PAT after post-processing, PATa.

Figure 10. The relationship between PPG signal quality and PAT with estimated matched-pairs rc values. Interquartile range of PAT across all MESA subjects in: (a) high- (SQI = 1) and low-quality (SQI = 0) PPG segments, and (b) regular and irregular rhythm PPG beat-to-beat intervals; (c) the proportion of high-quality PPG during regular and irregular rhythm PPG pulse waves. p < 0.001 and rc≥ 0.50 is marked ***.

Figure 11. An example of PAT variations in different sleep stages: (a) PPG signal; (b) recorded hypnogram during PSG; (c) PPG signal quality estimated by the ECG-guided algorithm, SQI (SQI = 1—high-quality, SQI = 0—low-quality); (d) PAT after post-processing, PATa, with annotated obstructive sleep apnea (OSA), hypopnea (HA), oxygen desaturation (OxyDes), and arousal (Ar) episodes.

Figure 12. The interquartile range of PAT across all MESA subjects with estimated matched-pairs rc values in: (a) different sleep stages; (b) different events related to sleep-disordered breathing. Results are shown before and after PAT post-processing, PATb and PATa, respectively. p < 0.001 and rc≥ 0.50 is marked ***.