1. Estimating Heart Rate using Wrist-Type Photoplethysmography and Acceleration Sensor while Running
34th Annual International Conference of the IEEE EMBS
San Diego, California USA, 28 August - 1 September, 2012
Hayato Fukushima, Haruki Kawanaka, Md. Shoaib Bhuiyan and Koji Oguri

2. Introduction
Monitoring heart rate during movement is important for preventing accidents caused by heart troubles.
This paper’s purpose is estimating heart rate while actual running is being occurred, using Photoplethysmography (PPG) wrist-type sensor.
The arm portion is greatly affected by movement artifacts, and the waveform of PPG is easy to be distorted by body motion artifacts.
Estimation heart rate in this paper is using accelerometer to improve the accuracy of heart rate measurement.

3. Photoplethysmography
PPG is the change of blood’s volume in the capillary vessel.
PPG can observed noninvasively with irradiating LED toward the capillary vessel in the skin, and receiving the reflected light at PD.

4. PPG is greatly affected by movement artifacts, the artifact caused by the runner shaking arm convolutes into PPG.
Such artifact is periodic, and the frequency of the artifact is close to that of heart rate.

5. Outline of Heart Rate Estimation Algorithm
In this study, defined the following three states (resting, warming up, running) from the magnitude of the measured acceleration amount.
In each of these three states, the time and force of arm movement is different, so the estimation method can use the acceleration’s integral absolute value (it has both positive and negative values) as a threshold to determine each of the above three states.

6. Estimation using Frequency Analysis
Accelerometer used to reject the artifact of shaking arm. The artifact caused by motion appears in acceleration as well as in PPG.
Frequency sequences Xppg (f), Xacc(f) are calculated by frequency analysis from PPG and acceleration respectively

7. The frequency of heart rate detected in the frequency sequence D(f)
The frequency of heart rate detected in the frequency sequence D(f). D(f) is the difference of spectrum between PPG to Acceleration, defined as
The frequency whose spectrum is maximum in D(f) is accepted as heart rate.
With this method, heart rate frequency fHR can be detected, and the artifact of shaking arms can also be rejected.

8. Method for Aperiodic Motion
While in aperiodic movement such as warming up, artifacts can’t be detected in frequency sequences Xppg (f), Xacc(f) as clear peaks.
The relative reliability to estimate is defined by measured acceleration. The reliability as coefficient is calculated by acceleration (i) which is acceleration at i, in seconds. Reliability (i) is defined as

9. Heart rate frequency fHR estimated with weighted smoothing the instantaneous heart rate frequency fpeaks (i)
The instantaneous heart rate frequency fpeaks (i) is calculated by the difference of the time PPG’s peaks happen.

10. Post-Processing
Smoothing filter at the postprocessing used to reject outliers, caused by motion artifacts having huge magnitude.
n = 120 second used to calculate σ(i) and fav(i) at the time i, in seconds. σ(i) is defined as
fav(i) is the average of heart rate frequency, and the standard deviation is σ(i). Heart rate frequency is f(i).

11. If the difference between f(i) to fav(i) is greater than σ(i), such f(i) is determined to be an outlier. Then, the frequency is updated to fav(i). This filter is non-linear.
The updated frequency fnew(i) through the filter which rejects the outliers is defined as
Finally, the ultimate frequency ffinal is defined as
ffinal is calculated from smoothing fnew (i).

13. EXPERIMENT
The wrist-type PPG sensor weared in left wrist. LED to measure PPG is built on the back of the wrist-type sensor, and the accelerometer is built-in.
Heart rate is defined as the number of times the heart beats in one minute. In practice, instantaneous heart rate calculated from the difference of time of beats is evaluated.
For comparison, test subjects also wore holter ECG sensor which is medical equipment.This holter ECG is able to detect R wave even during movement.

14. Result
The blue line shows the correct value calculated from holter ECG.
The method using acceleration (red line) in this paper has higher accuracy than the comparison value affected by artifacts (green line).
The result of all subjects is that correlation coefficient r = 0.98, and the standard deviation of error SD = 8.7 bpm.

15. CONCLUSION
This paper reports estimating heart rate using wrist- type PPG sensor while the test subject is actually running.
The proposed method has a higher degree of usability compared to existing methods using ECG.
As of the experiment, the results were closer to holter ECG in high accuracy (r = 0.98, SD = 8.7 bpm).