A Comparison of Photoplethysmography and Near Infrared Spectroscopy Measurements on Four Body Sites
Mahsa Khalili1*, Saud Lingawi2, Brian Grunau1, Babak Shadgan2, Calvin Kuo2
1Department of Emergency Medicine, University of British Columbia (UBC); 2School of Biomedical Engineering, UBC
Introduction: Near-infrared spectroscopy (NIRS) and photoplethysmography (PPG) are among the commonly
used optical sensors for non-invasive monitoring of cardiac function1. Specifically, PPGs are widely used in
wearable technologies, such as smart/sports watches. Although the performance of PPG sensors for the detection
of heart rate on differ ent body sites has been studied previously2, little is known about the accuracy of NIRS sensors
to detect heartbeats on different body parts. This study aimed to evaluate the accuracy of PPG and NIRS
measurements to detect heartbeats at 4 body sites and compare the perfor mance of these two sensor modalities.
Methods: The sensor setup consisted of a single-lead ECG (SparkFun Electronics AD8232), 3 PPGs (Maxim
IntegratedTM MAXREFDES117), and a 3-Channel NIRS (Artinis Medical Systems PortaLite mini). The data
acquisition system included a microcontroller board (Adafruit Industries Adalogger M0) and an I2C multiplexer
(Texas InstrumentsTM TCA9548A) for multi-PPG data collection. Synchr onized ECG and PPG data were sampled
at 50Hz and logged on the Adalogger’s SD card. NIRS data were streamed to a laptop via Bluetooth at 50Hz and
exported using Artinis Medical Systems’ OxySoft software. The experimental procedure consisted of 4 sensor
placement trials (Finger, Head, Ear, Forearm/Wrist), which involved collecting 2 minutes of continuous data from
all sensors. Participants were instructed to remain seated and avoid voluntary movements to reduce motion artifacts.
We used a 4th-order zero-lag Butterworth band-pass filter to filter ECG (0.5–24 Hz) and PPG/NIRS (0.5–8 Hz) data.
A cross-correlation analysis was implemented to synchronize PPG and NIRS measurements. We used a 100s subset
of all sensor data (10-110s) for data analysis. A modified Pan-Tompkins algorithm3 with customized peak-detection
procedures was used to detect ECG R-Peaks and PPG/NIRS systolic peaks. For each participant and sensor
placement trial, the nu mber of detected ECG peaks was used as ground-truth heartbeat data. For each trial, the
difference between the total number of detected PPG/NIRS and ECG peaks were calculated and divided by the total
number of ECG peaks (i.e., normalized heartbeat detection error). We performed non-parametric analysis of
variance to compare PPG and NIRS heartbeat detection errors across all trials (SPSS 27, IBM Corp).
Results and Discussion: 8 healthy individuals with no cardiac history (4 Females; Age: 25±5 years; Weight:
66.3±14.8 kg; Height: 170.6±9.0 cm) were consented under University of British Columbia Ethics H21-00535. The
percent error of detected heartbeats by PPG and NIRS sensors as well as sample measurements for different trials
are presented in Figure 1. From our analysis, the PPG locat ed on the finger had the lowest error overall, while the
NIRS on the forearm had the highest error. For each sensor placement, while differences were observed b etween
the distribution of the heartbeat det ection error of the PPG and NIRS sensors, these variations were not statistically
significant (Friedman's Two-Way Analysis of Variance, p-value > .05). Similarly, the differences between the
distribution error of each sensor modality (i.e., PPG and NIRS) were not significantly different across the 4 sensor
placement trials (Independent-Samples Kruskal-Wallis Test, p-value > .05).
Figur e 1- PPG/NIRS sensor placement (top); Percent heartbeat err or and sample PPG/NIRS measurements (bottom)
Conclusions: The findings of this study suggest that PPG and NIRS have variations in heartbeat detection error
across locations, indicating variations in signal quality. Future work with additional subjects will explore
quantifying this relationship.
1. Abay, T.Y. et al. (2014). Annu Int Conf IEEE Eng Med Biol Soc, 2014, 5361–5364
2. Longmore, S.K. et al. (2019). Sensors, 19, 1874
3. Pan, J. et al. (1985). IEEE Transactions on Biomedical Engineering, BME-32, 230–236