Perfusion Index and Heart Rate Estimation Variations: Analysis of Photoplethysmography (PPG) Sensor Measurements

Introduction:: Wearable devices are widely used for remote cardiac monitoring in clinical applications.¹ While Electrocardiogram (ECG)-based devices, such as Holter monitors or ECG chest patches are prevalent in short-term cardiac monitoring,² they are not suitable for prolonged use due to usability concerns (discomfort, skin irritation). Photoplethysmography (PPG) is an optical sensor for non-invasive monitoring of cardio-respiratory function, such as heart rate (HR), heart rhythm, and arterial oxygen saturation (SpO₂).³ Compared to ECG devices, wearable PPGs are more convenient for long-term use and are adaptable to different form factors (finger, wrist, ear).⁴ However, a limitation of PPG-based cardiac monitoring is that the signal quality can be affected by various factors, including local blood circulation, skin/tissue structure, and pressure contact.⁵ Hence, characterizing PPG signal variations is imperative for accurate interpretation of cardiac function.

PPG waveforms have two primary components, pulsatile (AC) and non-pulsatile (DC).^4,6 Perfusion index (PI), defined as the ratio of the PPG-AC over PPG-DC (PPG-AC/PPG-DC), has been used as a gold standard measure of PPG signal quality.^7,8 Comparison of PI variations on different anatomical sites has shown that fingers, face, and ears had the highest PI compared to other locations, such as the arms and legs.⁹ While it is generally accepted that low PI represents low signal quality and poor HR estimates, it is unclear how PI variations are quantitatively related to HR estimation errors. The main objective of this work was to investigate PPG waveform characteristics and HR estimation errors on different anatomical sites and in multiple measurement sessions.


Materials and Methods:: Experimental Setup. Single-lead ECG (AD8232), 3 PPGs (MAXREFDES117), and a microcontroller (Adalogger M0) were used to collect synchronized ECG-PPG data at 50 Hz. The 3 ECG electrodes were attached to the upper chest near both clavicles and below the right ribcage.

Experiment 1. PPG data were measured from 12 body sites: tip of the middle finger (Left: L, Right: R), the base of the index finger (L), wrists (L, R), forearm (R), forehead, temples (L, R), earlobes (L, R), and mastoid bone (R).

Experiment 2. Inter-/intra-session PPG measurement variations were analyzed. First, PPGs were attached to the base of the index finger and the tip of the middle finger (L). Participants remained seated and motionless while PPG-ECG data were collected for 2 min (Session 1). Next, both sensors were removed and re-attached and the measurements were repeated (Session 2a). Next, participants were instructed to make random hand/arm movements for 2 min. Following this, participants were instructed to remain stationary while PPG-ECG data were collected for 2 min (Session 2b).

Data Analysis. A 100s subset of measurements was used for analysis (10s-110s since start time). Band-passed ECG-PPG signals (4th-order-zero-lag Butterworth, ECG: 0.5–24 Hz; PPG 0.5–8 Hz) were used for peak detection. A modified Pan-Tompkins algorithm was used for ECG R-Peak and PPG systolic peak detection.¹⁰ ECG-/PPG-driven HR estimates were calculated, with ECG representing the ground-truth HR. Low-passed (cut-off: 8 Hz) PPG measurements were used for PPG-AC/-DC calculations. The local PI was derived by dividing PPG-AC over PPG-DC.


Results, Conclusions, and Discussions:: Results. We consented 8 individuals with no cardiac history (4 females, aged 25±5 years) (#H21-00535). Large variations were observed in PI and HR estimation error on different body parts (Fig.1), which were predominantly determined by PPG-AC amplitude variations (i.e., detectable pulse). No specific trend was observed in the PPG-DC component. Overall, measurements on the forearm/wrists had the lowest PI (< 0.2%) and highest HR estimation error (1.64±6.35 [mean%±std%]). Conversely, fingertips had the highest PI ( >1%) and the lowest HR error (−0.29±4.26).
PPG-DC, PPG-AC, PI, and HR estimation error variations, as well as inter-session (difference between PPG measurements in Session 1 and Session 2a), and intra-session (PPG measurement changes between Session 2a and Session 2b) variations were examined (Fig.2). The evidence suggests higher inter-/intra-session variations in PPG-DC compared to PPG-AC and PI. While small changes were observed in HR estimation error (< 2%), the average HR error remained close to zero (inter-session-variability:−0.38%; intra-session-variability:−0.57%).
Discussion. We investigated the relationship between PPG waveform variations and HR estimation error. Our findings support the use of PI to measure signal quality, but more importantly, to quantify the expected HR estimation error given an observed PI. Different PPG placements revealed that higher PI was associated with lower HR errors, with PI >1% having the lowest HR estimation errors (< 1%), and PI < 0.2% having HR errors as large as ±6%. The observed higher inter-/intra-session variability with PPG-DC compared to PPG-AC and PI are likely due to changes in sensor-to-skin contact pressure. Despite these variations, no significant change in PI or HR estimation error was observed. Overall, the findings support that PI variations and subsequently HR estimation error were predominantly determined by the pulse strength (PPG-AC).
Conclusion. PPG signal quality (PI) changes across different body parts that can result in HR estimation error variations at different locations. This work provided preliminary evidence that PI can be used to estimate HR accuracy, which can be an indicator of that location’s suitability for PPG application. Considering the popularity of PPG wearables in remote cardiac monitoring, the accuracy and trustworthiness of these measurements should be interpreted with caution.


Acknowledgements (Optional): :

References (Optional): : 1. Hilty, D. M. et al. Sensor, Wearable, and Remote Patient Monitoring Competencies for Clinical Care and Training: Scoping Review. J. technol. behav. sci. 6, 252–277 (2021).
2. Lobodzinski, S. S. ECG Patch Monitors for Assessment of Cardiac Rhythm Abnormalities. Progress in Cardiovascular Diseases 56, 224–229 (2013).
3. Almarshad, M. A., Islam, M. S., Al-Ahmadi, S. & BaHammam, A. S. Diagnostic Features and Potential Applications of PPG Signal in Healthcare: A Systematic Review. Healthcare (Basel) 10, 547 (2022).
4. Ghamari, M. A review on wearable photoplethysmography sensors and their potential future applications in health care. International Journal of Biosensors & Bioelectronics 4, (2018).
5. Reiss, A., Indlekofer, I., Schmidt, P. & Van Laerhoven, K. Deep PPG: Large-Scale Heart Rate Estimation with Convolutional Neural Networks. Sensors 19, 3079 (2019).
6. Tamura, T., Maeda, Y., Sekine, M. & Yoshida, M. Wearable Photoplethysmographic Sensors—Past and Present. Electronics 3, 282–302 (2014).
7. Elgendi, M. Optimal Signal Quality Index for Photoplethysmogram Signals. Bioengineering 3, 21 (2016).
8. Park, J., Seok, H. S., Kim, S.-S. & Shin, H. Photoplethysmogram Analysis and Applications: An Integrative Review. Front Physiol 12, 808451 (2022).
9. Tur, E., Tur, M., Maibach, H. I. & Guy, R. H. Basal Perfusion of the Cutaneous Microcirculation: Measurements as a Function of Anatomic Position. Journal of Investigative Dermatology 81, 442–446 (1983).
10. Pan, J. & Tompkins, W. J. A Real-Time QRS Detection Algorithm. IEEE Transactions on Biomedical Engineering BME-32, 230–236 (1985).