Examining Interbeat Interval Measurement Capability of Wearable Health-Monitoring Algorithms

要約

Photoplethysmography- (PPG-) based heart-rate (HR) monitors are becoming a real option for consumer use, not only for HR monitoring while exercising, but also for measuring other health and wellness indicators such as sleep quality, stress, sports recovery, and so on. PPG provides a more comfortable solution, as it does not require electrodes to be placed on the body. Most of these analyses require heart beat as well as the time interval between heart beats termed as interbeat interval (IBI). This paper examines the IBI measurement capability of Maxim Integrated's wearable health-monitoring algorithms. The evaluation covers different types of datasets collected from several subjects during several resting states, and it evaluates IBI measurement from PPG.

Introduction

The time interval between individual heartbeats, called interbeat interval (IBI), is a crucial health indicator. It is the primary input for heart-rate variability (HRV), which is used to assess heart health and autonomous nervous system health. It may also be used in a wide spectrum of applications, such as clinical practice, sleep quality measurement, and stress and recovery analysis 1.

ECG and PPG waveforms with IBIs illustrated. Figure 1. ECG and PPG waveforms with IBIs illustrated.

Maxim’s wearable health-monitoring algorithms produce four different time-domain and six different frequency-domain HRV metrics. The time-domain metrics are AVNN, SDNN, RMSSD, and pNN50. The frequency-domain ones are ULF, VLF, LF, HF, LF/HF, and TOTPWR 2.

The objective of the study discussed in this paper is to evaluate the IBI measurement accuracy of Maxim’s optical HR monitor. The comparison is performed against the Firstbeat Bodyguard 2 (BG2) wearable RR interval recorder. The reference device is reported to have 2.96ms mean absolute error (MAE) in RRI as compared to the standard electrocardiogram (ECG). Selection of the reference device is justified by the accuracy evaluation of the device. As Figure 1 illustrates, the interbeats calculated from ECG and PPG should amount to the same time interval as suggested in 5.

Method for Algorithm Evaluation

For our study, data collection was performed using a wrist-worn device equipped with Maxim’s optical sensor, which is capable of sensing multi-channel, multi-waveband optical data. The reference device was attached to the subject’s chest with ECG electrodes.

Since both devices were not turned on at the same exact moment, the streams of data were synchronized with each other by minimizing their mean absolute difference. Afterwards, to compensate for eventual time drifts between Maxim and Bodyguard 2 clocks, the data was split into intervals of approximately one minute and a new synchronization for each interval was performed. Details of this alignment can be found in a later section.

IBI evaluation covers only the resting state of the subjects since this measurement requires such. This is due to the nature of the human body and hemodynamics (the dynamic of blood flow). The PPG signal waveform is highly affected by the hemodynamics, which itself is highly affected by motion. In addition to the effect of hemodynamics, motion affects the interaction between the optical sensor and the skin. This interaction should ideally be an uninterrupted and undisturbed contact. Although the signal is conditioned and filtered by advanced signal-processing techniques inside Maxim’s wearable algorithms, IBI measurement may still be degraded by excessive motion. Therefore, the IBI module of the algorithm is restricted to be run in the absence of motion determined with a certain threshold on the accelerometer signal.

Data Collection

Collected data may be grouped into two categories: (1) short dataset and (2) one-hour dataset. The first category covers the following activities: sitting, walking, running, and resting for 15 minutes. In the second dataset, the subjects sat comfortably and watched TV (a program that did not induce any stress) for one hour, without sleeping.

The reference device used in this study is the Firstbeat Bodyguard 2 ECG-based HR monitoring device that records ECG with electrodes placed on the right upper body and left lower body. The measurement accuracy for RR recording is 1ms (sampling rate: 1000Hz).

Data Recording Conditions

  1. Activity Data Subjects were asked to perform the physical activities listed in Table 1. All activities were performed in a gym, using a treadmill for walking and running.
  2. Rest Data Rest data was collected during TV watching. The subjects were asked not to sleep. The data collection duration was limited to one hour.

Table 1. Activity Protocol for Data Collection

Activity Duration
Resting (Sitting) 5 min
Transition 1 min
Walking 5 min
Running 5 min
Walking 3 min
Resting (Standing) 2 min

Data Acquisition

  1. Short Dataset (Activity Protocol)
  • Data collection in gym
  • Average duration: ~15 minutes
  • Data has six channels: two green, two infrared, and two red
  • Data collection @25sps
  • Reference device is Firstbeat Bodyguard 2
  1. One-Hour Datasets (Watching TV)
  • Data collection at rest
  • Average duration: ~one hour
  • Data has six channels: two green, two infrared, and two red
  • Data collection @25sps

Performance Evaluation

Accuracy assessment of IBIs requires alignment between the reference measurement and the Maxim IBI measurement. Both the data from the Maxim algorithm and the data from the Bodyguard 2 reference device were processed with the Firstbeat artifact correction method 3. Ectopic beats were detected using the algorithm presented by Mateo et al. in 4 and excluded from the evaluation.

The following steps were taken for the accuracy assessment of the Maxim IBI:

  1. Find global shift between Maxim Integrated's and Firstbeat’s IBIs
  2. Split both in chunks of approximately 45 seconds
  3. Re-align chunks to minimize mean absolute error
  4. Calculate error and accuracy

IBI tachogram before (a) and after (b) alignment  (vertical axis is time in milliseconds, horizontal axis is the index number) Figure 2. IBI tachogram before (a) and after (b) alignment (vertical axis is time in milliseconds, horizontal axis is the index number).

Artifacts

HR time series artifacts are caused by several sources. They are common, and often characteristic, for healthy and clinical subjects, in both laboratory and field monitoring, from sleep to sports. In the measurement environment, magnetic, electric, and RF noise may disturb the device, especially HR monitors. Furthermore, the contact difficulties of electrodes, such as the lack of moisture, a problem in the measurement equipment, or spikes produced by body movements may trigger errors.

Also, internal artifacts that are initiated by the body exist. These arrhythmias are not actual artifacts in the technical sense, but are irregular in physiological terms, alter computations, and are, thus, treated as artifacts. Different instantaneous arrhythmias are normal also for healthy subjects and could be considered characteristic for ECG and the HR time series. Arrhythmias like tachycardia and bradycardia are pathological and may cause extra beats (EB) or missing beats (MB) in the corresponding RR intervals 4.

For automatically detected ectopic beats, both missing and extra beats are excluded from the subsequent error and accuracy calculation. Ectopic beats are accounted for additionally to the mean absolute error. Figure 3 and Figure 4 illustrate both cases where there are ectopic beats and there are none.

	Missing and extra beat illustrated on tachogram. Figure 3. Missing and extra beat illustrated on tachogram.

No missing or extra beat. Mean absolute error is below 2ms. Figure 4. No missing or extra beat. Mean absolute error is below 2ms.

Results

Results for TV Watching Datasets

IBI variance is highly affected by the physical and mental state of the body. In that regard, the TV watching dataset has some advantages compared to other datasets. Additionally, the wristband is more stable because the subject deliberately keeps his arm steady. The average error for this dataset is 6.77ms and reporting coverage is 95%. Extra beats (1.71%) and missing beats (0.86%) add up to 2.57% of total ectopic beats for this dataset. Details for each session can be found in Table 2.

Table 2. TV watching datasets

Dataset MAE(ms) Coverage(%) Extra(%) Missed(%)
1 6.86 95.52 1.89 2.59
2 4.42 99.55 0.29 0.16
3 3.49 99.76 0.16 0.08
4 3.63 100 0 0
5 6.88 97.28 1.98 0.74
6 6.21 97.44 1.87 0.69
7 11.68 91.42 6.39 2.19
8 9.35 96.98 1.7 1.32
9 10.05 95.3 3.58 1.12
10 5.21 99.24 0.49 0.27
11 6.72 99.25 0.49 0.26
Average 6.77 95 1.71 0.86

Results for Activity Protocol Datasets

Activity protocol datasets consist of several activities; however, the scope of the IBI evaluation presented here is limited to the resting state of the subject. The MAE averaged over 17 subjects amounts to 12.2ms, while the coverage is 68%, which leaves a total of 32% ectopic beats. Details for each session can be found in Table 3.

Table 3. Activity protocol datasets

Dataset MAE(ms) Coverage(%) Extra(%) Missed(%)
1 7 67.9 18.2 13.9
2 19.1 83.5 9.3 7.3
3 3.5 67.5 8.2 24.3
4 10.9 58.4 28.5 13.1
5 5.9 72.9 6.3 20.8
6 16.8 64.5 19.8 15.8
7 15.2 50.2 30.3 19.5
8 14.3 64.1 25 11
9 11 78.3 8.2 13.5
10 13.4 57.9 30.7 11.5
11 12.1 59.2 29.7 11.1
12 5.8 65.6 13.6 20.9
13 11 69.9 22.9 7.3
14 8.8 78.9 9.2 11.9
15 16.1 87.7 5.6 6.7
16 25.8 60.4 31.9 7.6
17 10.3 71.5 13.6 14.9
Average 12.2 67.9 18.2 13.9

Conclusion

The study presented in this paper evaluates the accuracy of IBI using a PPG-based, wrist-worn device. The Maxim wearable health-monitoring algorithm correctly detected more than 95% of the heart beats with an average IBI accuracy of approximately 6.77ms when motion was at a minimal level during TV watching. The results can vary by activity due to motion and sensor skin interaction.

The results demonstrate that new PPG-based HR monitors are becoming a real option for consumer use, not only for HR monitoring while exercising, but also for HRV analysis. PPG provides a more comfortable solution, as it does not require electrodes to be placed on the body.

References

  1. Parak, J., Tarniceriu, A., Renevey, P., Bertschi, M., Delgado-Gonzalo, R., & Korhonen, I. (2015, August). Evaluation of the beat-to-beat detection accuracy of PulseOn wearable optical heart rate monitor. In 2015 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 8099-8102). IEEE.
  2. Shaffer, F., Ginsberg, J. P., “An Overview of Heart Rate Variability Metrics and Norms”, Frontiers in Public Health, vol. 5, 2017.
  3. Saalasti S et al. (2004). Artefact Correction for Heartbeat Interval Data. Advanced Methods for Processing Bioelectrical Signals. Available: http://www.firstbeat.com/userData/firstbeat/download/ saalasti_et_al_probisi_2004_congress.pdf
  4. J. Mateo, P. Laguna, ”Analysis of heart rate variability in the presence of ectopic beats using the heart timing signal,” IEEE Transactions on Biomedical Engineering, vol. 50, pp.334-343, 2003.
  5. Pinheiro, N., Couceiro, R., Henriques, J., Muehlsteff, J., Quintal, I., Goncalves, L., & Carvalho, P. (2016, August). Can PPG be used for HRV analysis?. In 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 2945-2949). IEEE.