Professional Documents
Culture Documents
Abstract – A Photoplethysmographic (PPG) signal is invariably even slightest movement of patient disturbs the PPG signal,
corrupted with motion artifacts (MA) due to voluntary or resulting in inaccurate estimation of SpO2. Hence, reduction of
involuntary movements of the patient. PPG is a non-invasive motion artifact (MA) has been a challenging pulse oximetry
signal, used for the estimation of arterial blood oxygen saturation problem, ever since its invention, for a reliable estimation of
(SpO2), which helps the physician to know the hypoxic status of
SpO2.
patient during clinical investigations. This paper presents an
efficient Adaptive Step- size Least Mean Squares (AS-LMS) Signal processing techniques are efficiently used for MA
based adaptive filter for reducing the MA from corrupted PPG reduction from the corrupted PPG signals. Since the frequency
signals. The novelty of the proposed algorithm lies in the fact that spectrum of this noise (MA) overlaps frequency spectrum of
a synthetic noise reference signal for adaptive filtering, the desired signal (artifact-free PPG), traditional filtering
representing MA noise, is generated internally from the MA methods fail to remove the noise. To some extent, effect of
corrupted PPG signal itself instead of using any additional motion artifacts can be reduced by displaying the average
hardware such as accelerometer or source-detector pair for value of several SpO2 readings, which is a typical procedure
acquiring noise reference signal. Convergence analysis, SNR followed by the manufacturers of commercial pulse oximeters.
calculations and Statistical analysis revealed that the proposed
To deal with this challenging problem, several methods
AS-LMS algorithm has a clear edge over the Time-Varying Step-
size LMS (TVS-LMS) and Constant Step-size LMS (CS-LMS) were proposed for MA reduction. The most common
based adaptive algorithms for MA reduction from PPG signals. technique employed for the MA reduction is the moving
Experimental results, for the PPG data recorded with different average method [2]. But, it works well only for a limited range
motion artifacts (Horizontal, Vertical and Bending motion of of artifacts. Practically, the in-band noise due to MA
finger), demonstrated the efficacy of the proposed algorithm in corruption can be successfully reduced with the use of
MA reduction and thus making it best suitable for real-time pulse adaptive filters [3]-[5], by keeping in view of simplicity and
oximetry applications. advantages like faster response time, avoidance of pulse
Keywords- PPG, MA, AS-LMS Adaptive Filter. segmentation and ability to continue processing under MA
conditions. Use of a synthetic reference signal estimated from
I. INTRODUCTION the artifact-free part of the PPG signal [6] was also reported
Photoplethysmography, a noninvasive technique, is for reducing motion artifact. Significant improvements were
frequently used in clinical investigations for the measurement observed by applying time-frequency methods like smoothed
of heart rate and arterial oxygen saturation (SpO2). PPG is a pseudo Wigner-Ville distribution [7], making use of the non-
pulsatile waveform acquired from pulse oximeter using simple stationary nature of the PPG signals. In this paper, we present
inexpensive optical sensors (red and IR PPG sensors). The a simple and efficient adaptive technique, focusing on accurate
simplest PPG sensor consists of an infrared LED and a photo reconstruction of the signal waveform, which automatically
detector placed in a small plastic housing. The sensor is help in better estimation of SpO2, rather on attempting to
applied to the skin by means of a double-faced adhesive ring. stabilize saturation estimates. The proposed adaptive
The sensor can be either of transmitting type or reflecting technique utilizes AS-LMS algorithm for MA reduction. In
type. The pulsatile portions of red and IR PPG signals may be general, extra hardware like additional source-detector pair or
used for reliable estimation of accurate SpO2. The part of accelerometer are employed for generating a reference signal
detected PPG signal due to the arterial blood appears pulsatile for adaptive filtering application to PPG signals [3]-[5]. The
in nature at the heart rate. The pulsatile portion of the PPG novelty of the proposed method lies in the very fact that a
arises due to the light passing through arterial blood and hence synthetic noise reference signal is generated from the recorded
has information contained in the arterial blood flow like; heart PPG data itself.
rate, heart rate variability, respiration and blood pressure to
name a few [1]. Reliable and accurate estimation of SpO2 II. MOTION ARTIFACT REDUCTION
requires clean artifact-free PPG signals with clearly separable Adaptive filters are proved to be best filters in noise
DC and AC parts. The main cause for deterioration of cancellation by self adjusting the filter coefficients based on
accuracy in pulse oximeters is the corrupted portions created an adaptive algorithm, or close-loop adaptation. Different
in the detected PPG signals with voluntary or involuntary algorithms [8] may be employed to adapt the weights w of the
movements (motion artifacts) of patient. While in process, filter, with an attempt to minimize the Mean Square Error
-4
100
Magnitude
5
(c2)
Figure 3. Developed prototype PPG analog front-end (c1) 0
-5
5 100
10
(d2)
(d1)
-5 0
IR PPG 5 100
(e2)
(e1)
Amplitude (V)
-5 0
0 10 20 30 35 0 1 2 3 4 5
Time (S) Frequency (Hz)
Figure 6. IR PPG with “Horizontal motion of finger”: (a1) MA corrupted
RED PPG PPG (b1) Generated synthetic noise reference signal (c1) MA recovered PPG
using CS-LMS (d1) MA recovered PPG using TVS-LMS (e1) MA recovered
-10
0 20 40 60 80 100 120 140 160 PPG using AS-LMS and their corresponding spectra in (a2)-(e2)
Time (S)
3 400
Figure 4. MA corrupted PPG data recorded from a volunteer
(a2)
0
-3 (a1) 0
Constant Step Size LMS (CS-LMS)
-100
Time Varying Step Size LMS (TVS-LMS) 1 200
Adaptive Step Size LMS (AS-LMS) (b2)
(b1)
-200 0
Mean Squre Error (dB)
-1
Amplitude (V)
1 100
Magnitude
-300
(c2)
-400 -1 (c1) 0
1 200
-500
(d2)
(d1)
-600 -1 0
1 200
-700
0 200 400 600 800 1000 0 (e2)
No. of Iterations
-1 (e1) 0
Figure 5. Mean Square Error against the number of Iterations 0 20 40 0 2 4
Time (S) Frequency (Hz)
With thorough experimental analysis on the recorded IR
PPGs, simulated output results along with their corresponding Figure 7. IR PPG with “Vertical motion of finger”: (a1) MA corrupted PPG
spectra (for typical motions of finger like Horizontal, Vertical (b1) Generated synthetic noise reference signal (c1) MA recovered PPG using
CS-LMS (d1) MA recovered PPG using TVS-LMS (e1) MA recovered PPG
and Bending motions) before and after MA reduction are using AS-LMS and their corresponding spectra in (a2)-(e2)
portrayed in Fig. 6 to Fig. 8. The recovered portions of MA
3 500 (AC Re d /DC Re d )
(a2) R= (9)
(AC IR /DC IR )
(a1) 0
-4 Using R of (9), SpO2 is calculated using the relationship
2 400
(b2) SpO2 % = (110 − 25R) % (10)
-2 (b1) 0
Amplitude (V)
Magnitude
200
RECOVERED PPG FOR DIFFERENT MOTIONS
(c2)
Horizontal Vertical Bending
-1 (c1) 0 motion motion motion
PPG
1 200
(d2) R SpO2 R SpO2 R SpO2
(d1) 0
-2 Clean 0.68 0.720 0.56
2 200 93̃ 92̃ 96̃
section ±0.048 ±0.059 ±0.13
(e2)
(e1) Using 0.70
92.̃
0.724
91.̃
0.6
95̃
-2 0 CS-LMS ±0.035 ±0.055 ±0.06
0 20 40 0 2 4
Time (S) Frequency (Hz) Using 0.69 0.722 0.58
92.7.̃ 91.9.̃ 95.5.̃
TVS-LMS ±0.032 ±0.051 ±0.07
Figure 8. IR PPG with “Bending motion of finger”: (a1) MA corrupted PPG
Using 0.69 0.722 0.58
(b1) Generated synthetic noise reference signal (c1) MA recovered PPG using 92.7.̃ 91.9.̃ 95.5.̃
AS-LMS ±0.029 ±0.047 ±0.07
CS-LMS (d1) MA recovered PPG using TVS-LMS (e1) MA recovered PPG
using AS-LMS and their corresponding spectra in (a2)-(e2)
IV. CONCLUSION
A. SNR Calculation
Reliable and accurate measurement of SpO2 using pulse
A figure of merit, SNR, defined as the ratio of signal oximeter is highly influenced by MA introduced in to the PPG
power to the generated noise reference power is computed for signals. Reduction of MA from PPG signals using AS-LMS
the MA corrupted and MA recovered PPG signals. Table II adaptive filtering technique is presented in this paper. The
illustrates the SNR values for the PPGs inflicted with three novelty of the method lies in the fact that a synthetic noise
different kinds of MA. The superiority of AS-LMS is clearly reference signal, representing the MA, is generated for use in
evident in representing high SNR values in all the cases. adaptive filtering without using additional hardware. Fast
convergence of AS-LMS compared to both TVS-LMS and
TABLE I. EFFECTIVENESS OF PROPOSED METHOD IN RESTORING
PEAK-TO-PEAK VALUES OF PPG
CS-LMS techniques, in addition to restoring the peak-to-peak
Horizontal Vertical Bending values of the PPG cycles by it, resulted in efficient reduction
PPG of motion artifacts from the PPG signals. Test results on the
motion motion motion
Clean 0.371 0.428 0.357 recorded PPG signals resulted in a reliable estimation of SpO2
Section ± 0.025 ± 0.045 ± 0.022 values, thus making the proposed method best suitable for
0.426 0.514 0.459 reduction of MA in pulse oximeter.
Corrupted portion
± 0.087 ± 0.107 ± 0.067
REFERENCES
Recovered portion using 0.388 0.443 0.364
CS-LMS ± 0.048 ± 0.059 ± 0.131 [1] A. B. Hertzman, “The blood supply of various skin areas as estimated by
Recovered portion using 0.379 0.435 0.359 photoelectric plethysmograph,” Am. J. Physl., vol.124. pp. 328-340,
TVS-LMS ± 0.036 ± 0.048 ± 0.128 1938.
Recovered portion using 0.379 0.435 0.359 [2] T. L. Rusch, R. Sankar and J. E. Scharf, “Signal processing methods for
AS-LMS ± 0.031 ± 0.046 ± 0.125 pulse oximetry,” Comput. Biol. Med., vol. 26, no. 2, pp. 143- 159, 1996.
[3] B. Barreto, L. M. Vicente and I. K. Persad, “Adaptive cancellation of
motion artifact in photoplethysmographic blood volume pulse
TABLE II. COMPUTED VALUES OF SNR FOR MA CORRUPTED AND measurements for exercise evaluation,” in Proc. IEEE-EMBC and
MA RECOVERED PPG FOR DIFFERENT MOTIONS CMBEC, vol. 2, pp. 983-984, 20-23 sept.,1995.
Horizontal Vertical Bending [4] R. Relente and L. G. Sison, “Characterization and adaptive filtering of
SNR of PPG motion motion motion motion artifacts in pulse oximetry using accelerometers”, in Proc. Conf.
EMGS/BMES, Houston, TX, USA, pp. 1769-1770, 23-26 Oct., 2002.
Before Adaptive [5] K. W. Chan and Y. T. Zhang, “Adaptive reduction of motion artifact
1.2807 0.2686 0.2666
Filter from photoplethysmographic recordings using a variable step-size LMS
Using CS-LMS 1.5497 0.3591 0.3752 filter,” in Proc. IEEE Sensors, vol. 2, pp. 1343- 1346, 2002.
[6] F. M. Coetzee and Z. Elghazzawi, “Noise-resistant pulse oximetry using
Using TVS-LMS 1.7325 0.3888 0.3934
a synthetic reference signal,” IEEE trans. Biomed. Engg., vol. 47, no. 8,
Using AS-LMS 1.7334 0.3898 0.3936 pp. 1018-1026, 2000.
[7] Y. S. Yan, C. C. Poon and Y. T. Zhang, “Reduction of motion artifact in
B. SpO2 Computation pulse oximetry by smoothed pseudo Wigner-Ville distribution,” J
Neuroengineering. Rehabil., 2005, 2-3, available from
An artifact-free PPG signal is composed of clearly separable http://www.jneuroengrehab.com/content/2/1/3.
AC and DC parts. For the measurement of SpO2, the DC parts [8] V.K. Madisetti, D. V. Williams, “Digital Signal Processing Hand Book,”
of the red (DCRed) and Infrared (DCIR) PPG signals are CRC press, 1999.
extracted. Similarly, the peak to peak values of the pulsatile [9] K. A. Reddy, B. George, N. M. Mohan and V. J. Kumar, ”A Novel
Calibration-Free Method for Measurement of Oxygen Saturation in
components at the heart rate ACRed and ACIR of the red and IR Arterial Blood”, IEEE Trans on Instrumentation and Measurement,
PPG signals respectively are measured and the “ratio of ratios” Vol. 58, no. 5, pp 1699-1705, May 2009.
R is estimated [9] as