We recruited 40 healthy participants between 18 and 40 years old with no current medical conditions and who were not taking regular medications. Demographics characteristics were collected. Body height and weight were measured under standard anthropometry procedures, and body mass index was calculated as weight in kilograms divided by height in square meters. Blood pressure was measured using an automatic blood pressure monitor (Tango M2, SunTech Medical, Inc., NC, USA) before and after testing procedures. The participants were evaluated for facial skin color by using the von Luschan skin color chart (range 1-36) [25]. This study was approved by the Joint Chinese University of Hong Kong – New Territories East Cluster Clinical Research Ethics Committee (CREC Ref. No. 2016.550).

We set up 2 iPhones (iPhone 6S; Apple, Inc, Cupertino, CA, USA) and a 12-lead ECG treadmill (GE Series 2000, GE Medical Systems Information Technologies Inc, Milwaukee, WI, USA) for HR measurements. Backdrop and background light intensity was standardized during signal acquisition and was captured in the unit of lux. The Cardiio (Cardiio Inc, Cambridge, MA, USA) smartphone app was installed in the 2 iPhones for facial and fingertip PPG detection. HR measurements were taken by continuous 12-lead ECG, and facial and fingertip PPG detection simultaneously. Each participant was given at least a 5-minute rest interval before testing began. We took 3 measurements of resting HR and averaged them for analysis before exercise. Participants were then tested on a motorized treadmill using the Bruce test protocol from stage 1 to stage 4 (2.7, 4.0, 5.5, and 6.8 km/h) [26]. Each participant underwent 2 sequential tests on the treadmill to achieve (1) moderate-intensity exercise, defined as 50% to 70% maximum HR, and (2) vigorous-intensity exercise, defined as 70% to 85% maximum HR. Maximum HR was calculated as 220 beats/min minus the participant’s age [27]. HR recordings were acquired immediately when moderate- and vigorous-intensity exercise was achieved based on HR from the continuous ECG and repeated successively until return to resting HR.

We present numerical results as mean (SD) or median (interquartile range, IQR). Pearson correlation (r) and coefficient of determination were applied to determine the nature of the relationship between the HR measurements made by the Cardiio smartphone app (face and fingertip PPG detection) and by the standard 12-lead ECG. We performed the paired Student t test and Wilcoxon signed rank test to determine the difference between the means and medians measured by both devices. Root mean square error (RMSE) was performed to evaluate the spread of errors between predicted and observed values. Bland-Altman plots were used to examine agreement between ECG and PPG-estimated HR. All statistical analyses were performed using IBM SPSS statistical software (IBM SPSS Statistics for Windows, version 22.0; IBM Corporation). All analyses were 2-tailed, and P values of <.05 were considered statistically significant.

We recorded a total of 665 fingertip and 665 facial PPG measurements. Boxplots were used to determine extreme outliers for any observation outside the upper and lower fences that were 3 times the IQR. Potential outliers were checked for accuracy before exclusion. Failed measurements (n=4 facial measures) and extreme outliers were possibly caused by the monitor losing fingertip skin contact or misalignment of the face. We excluded 69 outliers for PPG measurements representing unrealistic HR values (n=34 fingertip measures and n=35 facial measures), resulting in 631 fingertip and 626 facial PPG measurements for analysis.

Table 1 summarizes participants’ characteristics. Of the 40 healthy participants, 50% (n=20) were men, and the mean (SD) age was 24.7 (5.2) years. The median von Luschan skin color among participants was 23 (IQR 19-25); the lightest skin color was 14 and the darkest was 27. Male participants had a higher body mass index, resting systolic blood pressure (SBP), and resting diastolic blood pressure (DBP) than the female participants. There were no significant differences in postexercise SBP and DBP between the sexes. The median backdrop and background light intensity during signal acquisition was 199 lux (IQR 127-249).

We analyzed 80 averaged resting HR values from 234 PPG measurements, and overall 397 postmoderate- and postvigorous-intensity exercise HR values.

For resting HR, the Cardiio app captured 6% (5/80) of HR values ≥100 beats/min, and the mean (SD) HR was 73.41 (12.60) beats/min (range 52.67-112 beats/min) (Table 2). The mean (SD) difference in HR between fingertip PPG measurements and the reference ECG was –0.05 (1.03) beats/min (P=.69, paired t test; Table 3).

The mean (SD) HR of postmoderate-intensity exercise was 89.97 (19.95) beats/min (range 50-145 beats/min) with 28.2% (50/177) of HR values ≥100 beats/min captured by the Cardiio app. For postvigorous-intensity exercise, the mean (SD) HR was 105.12 (19.91) beats/min (range 57-157 beats/min) with 60.5% (133/220) of HR values ≥100 beats/min captured (Table 2). The mean (SD) differences in HR between fingertip PPG measurements and the reference ECG were –0.36 (2.14) beats/min (P<.05) for postmoderate-intensity exercise and –0.18 (2.02) beats/min (P=.19) for postvigorous-intensity exercise (Table 3).

Table 4 shows correlation coefficients of HR between the Cardiio smartphone app and the reference ECG. Fingertip PPG-estimated HRs were strongly correlated with ECG HR at rest (r=.997, P<.001; RMSE=1.03 beats/min or 1.40%), postmoderate-intensity exercise (r=.994, P<.001; RMSE=2.15 beats/min or 2.53%), and postvigorous-intensity exercise (r=.995, P<.001; RMSE=2.01 beats/min or 1.93%), as Figure 2 illustrates.

Figure 3 presents the Bland-Altman plots with 95% limits of agreement. The Cardiio smartphone app had 95% of resting HR measurements fall within –1.98 and +2.07 beats/min, while postexercise HR measurements were within –3.81 and +4.32 beats/min.

We analyzed 80 averaged resting HR values from 237 facial PPG measurements and a total of 389 postexercise HR values.

The mean (SD) HRs were 73.17 (12.64) beats/min (range 51.50-112.67 beats/min) at rest, 89.35 (19.47) beats/min (range 52-138 beats/min) postmoderate-intensity exercise, and 102.53 (19.16) beats/min (range 59-158 beats/min) postvigorous-intensity exercise (Table 2). Facial PPG detection determined 6% (5/80) of resting, 29.8% (56/188) of postmoderate-intensity exercise, and 54.7% (110/201) of vigorous-intensity exercise HR values ≥100 beats/min (Table 2).

The mean (SD) difference in HR between facial PPG detection and the reference ECG at rest was –0.29 (1.03) beats/min (P<.05), while postmoderate- and postvigorous-intensity exercise HR differences were –1.09 (3.67) beats/min (P<.05) and –0.93 (3.84) beats/min (P<.05), respectively (paired t test comparing fingertip PPG detection with reference ECG, Table 3).

The HR values from facial PPG measurements and the reference ECG were positively correlated (Table 4). The correlation of facial PPG-estimated HR with ECG HR (r=.997, P<.001; RMSE=1.02 beats/min or 1.44%) was strong at rest. Facial PPG-estimated HR was also strongly correlated with ECG HR after exercise under moderate intensity (r=.982, P<.001; RMSE=3.68 beats/min or 4.11%) and vigorous intensity (r=.980, P<.001; RMSE=3.84 beats/min or 3.73%) (Figure 2).

The Cardiio smartphone app had 95% of resting HR measurements fall within –1.72 and +2.30 beats/min, while postexercise HR measurements were wider and within –6.36 and +8.37 beats/min, as illustrated in the Bland-Altman plots (Figure 3).

According to the literature, HR monitors are considered accurate and regarded as excellent when r ≥.93 and RMSE<6.8% [28,29]. The American National Standards Institute/Association for the Advancement of Medical Instrumentation EC-13 standard states that the accuracy requirements for HR monitors are RMSE≤5 beats/min or ≤10%, whichever is greater [30]. Our results showed that the Cardiio smartphone app using both fingertip and facial PPG detection can be regarded as excellent and considered accurate in measuring HR at rest, and after moderate- to vigorous-intensity exercise.

PPG estimation of resting HR from both the fingertip and the face demonstrated very high accuracy. The RMSE were less than 1.5% (RMSE=1.40% from fingertip and RMSE=1.44% from facial PPG measurements). The errors between PPG-estimated resting HR and ECG HR were <1%, as illustrated by the coefficients of determination (Figure 2).

In particular, the results of fingertip PPG measurements were consistent and accurate in the different physical activity levels tested in this study. Fingertip PPG-estimated HRs were strongly correlated with both resting and postexercise ECG HR (all r ≥.99; Table 4). HR was underestimated with mean differences of <0.5 beats/min or median differences of 0 beats/min at rest and after exercise compared with the reference ECG (Table 3).

For facial PPG measurements, the accuracy of HR estimation was better at rest and diminished with exercise. The coefficient of determination of facial PPG-estimated HR was stronger with resting ECG HR (R²>.99) than postexercise ECG HR (R²≥.96) (Table 3). For postexercise HR detection, facial PPG underestimated HR values with mean differences of approximately 1 beat/min or median differences of 1-2 beats/min when compared with ECG HR (Table 3). Possible explanations are blushing and excessive facial motion due to heavy breathing after exercise leading to misalignment of the face with the camera.

Overall, the Bland-Altman plots showed better agreement between ECG and fingertip-estimated HR than between ECG and facial PPG-estimated HR, and that 95% limits of agreement were wider for facial than for fingertip PPG estimation (Figure 3). The results demonstrated the ability of a PPG-based smartphone app to provide meaningful and accurate readings at rest and after exercise.

Limitations of our study include that (1) this was a convenience sample of healthy young adults, which may limit generalizability; and (2) the method of HR detection from fingertip and facial PPG using iPhone’s front and back cameras, which may not apply to other smartphone cameras. We recommend further evaluation of the accuracy of taking HR measurements in a real-world environment under nonstandardized background lighting and in expanded exercise settings. We suggest conducting further studies including a larger number of participants with an extended age range and more skin tone colors to increase generalizability. This study did not include patients with arrhythmia or other heart-related problems, which is an area for future investigation. Future work could also study user feedback relating to health-sensing mobile apps.

Smartphone use is becoming ubiquitous. People are increasingly relying on smartphones and health-related apps for health care purposes. Smartphone apps must be validated for their accuracy, reliability, and effectiveness in providing health care benefits. Capturing a PPG signal with the built-in smartphone cameras may provide a readily accessible and inexpensive means to measure HR. Our results have demonstrated that HR detection by the Cardiio smartphone app is accurate at rest and after moderate- and vigorous-intensity exercise in a healthy young adult sample. Both fingertip and facial PPG have shown high accuracy in measuring resting HR with reference to an ECG. Although touchless facial PPG detection is more convenient, fingertip PPG is more accurate for HR detection after exercise.