ARTICLE IN PRESS. Computers in Biology and Medicine

Transcription

1 Computers in Biology and Medicine 39 (2009) Contents lists available at ScienceDirect Computers in Biology and Medicine journal homepage: Adaptive threshold method for the peak detection of photoplethysmographic waveform Hang Sik Shin, Chungkeun Lee, Myoungho Lee Department of Electrical and Electronic Engineering, Yonsei University, 134 Sinchon-dong, Seodaemun-gu, Seoul, Republic of Korea article info Article history: Received 29 August 2009 Accepted 8 October 2009 Keywords: Photoplethysmography (PPG) Peak detection Adaptive threshold Frequency analysis Respiratory noise abstract Photoplethysmography (PPG)-based temporal analyses have been widely used as a useful analytical method in physiological and cardiovascular diagnosis. Most of temporal approaches of PPG are based on detected peak points, peak and foot of PPG. The aim of presented study is the development of improved peak detection algorithm of PPG waveform. The present study demonstrates a promising approach to overcome respiration effect and to detect PPG peak. More extensive investigation is necessary to adapt for the cardiovascular diseases, whose PPG morphology has different form. & 2009 Elsevier Ltd. All rights reserved. 1. Introduction Corresponding author. Tel.: ; fax: addresses: glority@yonsei.ac.kr, micon78@yonsei.ac.kr, mhlee@yonsei.ac.kr (M. Lee). Nowadays, photoplethysmography (PPG) is widely used in cardiovascular and hemodynamic analysis [1]. PPG measures blood volume changes at a peripheral artery such as finger, toe, ear and forehead, and measured waveform has a little difference according to where it measured [2 4]. Though the different waveform is measured by measuring site, PPG waveform has a bottom or foot (V min ) and top or peak (V max ) points in common. V min represents minimum blood volume changes which corresponds beginning of ventricular contraction and blood ejection, on the other hands V max describes the maximum blood volume changes which means the end of blood ejection. PPG is generated by blood pressure and flow; however, it is an arbitrary unit signal because PPG is easily affected by environmental conditions not only sensor fitting method, but also skin condition, skin depth, race, humidity and circumference brightness. From these characteristics, it is hard to analyze using PPG amplitude, and PPG analysis has mainly carried out with a timing analysis and amplitude variability. The most of PPG applications have been intensively studied with temporal analysis such as pulse transit time (PTT) and pulse wave velocity (PWV). PTT and PWV mean the time taken and the speed, respectively, in the arterial pulse pressure wave to travel from the aortic valve to a peripheral site. PTT and PWV are widely used for physiological estimations such as arterial stiffness [5,6], left ventricular ejection time (LVET) [7], left ventricular preejection period (LVPEP) [8,9] and blood pressure [10,11]. With the development of temporal analysis based PPG applications, peak detection methods of PPG have become not special but very important issue in PPG temporal analysis because most of temporal analysis depend on peak position. PPG has a lesser sophisticated morphology than other physiological signals and this also means peak detection of PPG relatively easy because there are few specific points. However, PPG could have an enormous baseline drift and wondering followed by physiological condition and movement, moreover it frequently happens. It was demonstrated that PPG contains fluctuation caused by respiratory and sympathetic activity [12], even arousal changes such as drowsiness causes PPG baseline wandering or drift. These artifacts could be explained with the three major interferences of PPG, motion artifact, respiration effect and low perfusion. Motion artifact generally induces baseline drift; however, it also could be a reason of amplifier saturation which makes waveform loss. Respiration changes not only heart rhythm but also thoracic pressure. Rhythmic change followed by respiration is already known as respiratory sinus arrhythmia [13], and thoracic pressure also could have an effect on physical heart activity [14]. Therefore, PPG naturally contains a respiratory component, and it is reflected on the baseline and signal amplitude. With the low perfusion, which means weak blood flow of arterial blood to a capillary bed, these artifacts should be defined and removed for better detection. Most of previous researches were performed with maximum or minimum values detection of PPG waveform in classifying V max and V min by detecting local maxima or minima detection method (LCM) [15,16]. It is general acceptance that PPG is composed with incident wave and reflected wave [17 25]. Considering on respiration effect, the changes of reflected wave should be /$ - see front matter & 2009 Elsevier Ltd. All rights reserved. doi: /j.compbiomed

2 1146 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) investigated. The amplitude and velocity of reflected wave could be varied by respiratory activity, and it causes rapid changes of PPG waveform [26]. It was demonstrated that the reflected wave velocity becomes faster by increasing of age and vessel stiffness [21,27], and it was appeared in PPG [28,29]. In other words, we should discriminate the maximum position of incident wave from reflected wave disturbances for more precise analysis. These changes could be a reason of miss-detection of V max which closely related to reflected wave. LCM method is hard to apply for rapid changes of PPG waveform and different heart rate, and it also contains time delay from its specific window-size. Since PPG is frequently analyzed with ECG, in most cases, PPG peak detection has depended on ECG-gated method. However, these attempts are hard to use in the clinical application as well as it is increasing to use PPG as a single reference such as respiration and heart rate estimation [30]. The aim of the present study is summarized signal conditioning of PPG waveform and detection of V max and V min of PPG waveform. Frequency analysis based filtering was used for signal conditioning, and adaptive threshold (ADT) method was developed for peak point detection. ADT method was able to detect both V max and V min and evaluated by clinical experiment. 2. Methods 2.1. Subject Eighteen young and healthy subjects (11 male and seven female, mean ages of 24.1 years, range years, mean BMI of 23.2, range ) were participated in experiment. The subjects were normotensive (mean systolic/diastolic blood of 118/7076.3/5.6mmHg, range 100/63 140/8mmHg), and had no known cardiovascular, neurological or respiratory disease. Prior to the experiment, the subjects were requested to provide information about their physical condition. Physical information such as height and weight are also measured for demographical research and summarized in Table 1. Every experiment was performed in a typical sports medicine laboratory at an ambient room temperature from 11 am to 6 pm. Drinking and smoking were prohibited during 24 and 2 h before experiment, respectively. Table 1 Demographic data for the subjects in this study. Subject (male/female) Age (years) Height (cm) Weight (kg) BMI (kg/m 2 ) SBP a (mmhg) DBP b (mmhg) HR (bpm) c 1 (M) (M) (M) (M) (M) (F) (F) (F) (M) (M) (M) (F) (M) (F) (M) (M) (F) (F) Male (range) (17 30) ( ) (63 103) ( ) ( ) (66 80) ( ) Female (range) (19 26) ( ) (47 63) ( ) ( ) (63 75) ( ) Mean7SD (range) (17 30) ( ) (46 103) ( ) ( ) (63 80) ( ) a Systolic blood pressure. b Diastolic blood pressure. c Heart rate (beat per minute).

3 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) Signal measurement ECG and PPG were measured simultaneously by MP150 (Biopac, USA), ECG 100C and PPG 100C module. ECG was measured with LEAD II configuration on chest surface; PPG was measured on left index finger by TSD100B, plethysmography transducer. Acknowledge TM (BIOPAC, USA) was used for realtime monitoring and storing. For signal conditioning and peak detection, MATLAB 2008b (The MathWorks, Inc., Natick, MA, USA) was used. Omron HEM-907 was used for blood pressure measurement Experimental protocol Experiment was carried out in both the supine position and the sitting position, and signals were measured with spontaneous and controlled respiration to investigate respiration effect. Respiration was controlled with 0.1 (low rate) and 0.25 Hz (high rate) in the supine position by using metronome. In the sitting position, there was no respiration controlling. Experiment was sequentially performed with spontaneous, 0.1 Hz controlled and 0.25 Hz controlled respiration in the supine position, spontaneous respiration in the sitting position in order. Subjects had 2 min resting period between every experiments. Every data has 5 min length with 1 khz sampling rate PPG frequency analysis and filtering Frequency analysis of PPG waveform is very important because most of filtering methods are based on frequency characteristic of signal. There are few reports regarding the frequency characteristics of PPG; however, we could expect that PPG is composed with the major flow component and harmonics from the steadystate oscillation characteristics of arterial system [31]. We assume that the major frequency component of PPG is based on average pulse rate around 1 Hz, and harmonic frequencies are added up to construct PPG waveform. Low frequency component (o0.5 Hz) generally reflects the respiration ( Hz) and motion artifact (o0.1 Hz) Peak detection algorithm Bandpass filtering method could remove baseline distortion; however, severe morphological distortions from the low perfusion and baseline drift are hard to remove with frequency filtering. Additional filtering or feature extraction method such as moving average filter and wavelet decomposition could be useful to regulate signal, but filtering procedure should be minimized because most of PPG analysis are based on exact peak position which means little time delay. Proposed peak detection procedure consists of two parts, adaptive threshold detection method and peak correction Adaptive threshold detection In adaptive threshold detection, it was designed the virtual threshold which controlled by original PPG waveform amplitude. To detect foot or peak position, virtual threshold was increased or decreased with fixed slope parameter. In a V max example, the value of threshold is decreased by the passage of time during threshold is higher than the original signal amplitude. If threshold met the signal, threshold is accompanied with signal amplitude until it reaches to inflection point (V max ). After ADT finds V max, threshold is decreased again by a modified slope parameter. Former procedures are repeated until every peak is found. In V min detection, detection threshold is increased until threshold meets the signal, and is decreased by pre-defined slope parameter. Initial slope parameter and modified slope parameter were decided by (1) and (2), respectively, ( Slope Init ¼ 0:2argmaxðPPGÞ; V max detection ð1þ 0:2argminðPPGÞ; V min detection ðv n 1 þstd PPG Þ Slope k ¼ Slope k 1 þs r ð2þ F s In (2) Slope k, s r, V n 1, std PPG and F s mean k-th slope amplitude, slope changing rate, previous peak amplitude, standard deviation of entire PPG signal and sampling frequency, respectively. Slope changing rate is empirically selected as a 0.6 and 0.6 for a V max and V min. The period is defined as in-sensing period when threshold has fixed slope. On the contrary, out-of-sensing period means that detection threshold is accompanied with original signals. ADT procedure is graphically described in Fig. 1(a) Peak correction PPG waveform morphology could be varied with the relationship of incident wave and reflected wave. Cardiovascular characteristic which comes from aging, hypertension and vessel compliance or unwanted baseline changing have an effect on wave reflection and it could enhance or reduce reflected wave amplitude. V max Refractory period V max detection threshold V min detection threshold Ignored V min In sensing period Out of sensing period Fig. 1. (a) Adaptive threshold detection method. Solid line is bandpass filtered PPG waveform, and dashed line is detection threshold. V max and V min are described with. and m, respectively. In-sensing period, detection threshold is accompanied with PPG waveform amplitude, and in out-of-sensing period it is varied by pre-defined slope parameter. At upper and lower inflections, detection threshold reports the peak and it is only valid when peak is detected on out of refractory period. (b) If detected peak is found in refractory period, detected peak is ignored (r) and threshold level is not affected by ignored peak.

4 1148 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) Refractory period was adaptively selected by 0.6 times of pervious average pulse interval. Initial refractory period set 0.6 s and refractory period was calculated from previous peak position. Peak cancellation by refractory period is shown in Fig. 1(b) Evaluation For evaluation, true positive (TP), false positive (FP) and false negative (FN) were calculated. Then, we derived sensitivity (SE), positive predictivity (+P) and failed detection rate (FDR) to evaluate detection algorithm accuracy. Detected peaks were verified with 100 ms approval range from reference peaks. SE, +P and FD were calculated by (3) (5), respectively. Evaluation was carried out for both V max and V min detection. Detection performance of the proposed method was compared with the result of LCM. For a LCM method, moving window length and threshold level was decided by Fs/4 (The Fs refers to the sampling frequency) and Tha(i)=((i 1th) local minimum)/3, respectively, where Tha(1) is defined as the mean of the first 5 local minima, where i=1, 2, 3,y,N. Sensitivity ðseþ¼ Positive predictivity ðþpþ¼ Failed detection rate ðfdrþ¼ True positive True positiveþfalse negative True positive True positiveþfalse positive Failed detection Number of peaks ð3þ ð4þ ð5þ 3. Results 3.1. PPG signal conditioning Fig. 2 represents the frequency analysis of PPG using fast Fourier transform (FFT) of PPG during 0.25 Hz respiration (subject 11), in this case average heart rate (HR) was 71.6 bpm. In Fig. 2(a), it was found that frequency of PPG contains the major component at 1.2 Hz (72 bpm) around and harmonic components. These results are similar to previous result which about frequency characteristic of pressure and flow in a peripheral artery [31], and it was demonstrated the PPG reflects the peripheral artery blood flow. Fig. 2(b) represents the low frequency components indicated by a dotted box of Fig. 2(a). Controlled respiration was found at 0.25 Hz and baseline noise component was also found in the DC to 0.1 Hz range. Low frequency component of PPG contains respiration effect and baseline movement, thus Hz bandpass filter was used in signal conditioning. Because each PPG frequency components have high separability, discrete cosine transform (DCT) based filtering method was used for noise cancellation [32]. Though PPG DC component was regarded as an important part for some application, it usually reflects baseline changes which is hardly related to peak positions. Therefore, PPG DC component could be removed in peak detection and temporal analysis of PPG. Fig. 3(a) is noisy PPG and Fig. 3(b) represents bandpass filtered PPG waveform. From this Fig. 3(b), it was found that baseline wandering was removed, but low perfusion distortion still remained Y (f) Y (f) Hz 72 beats/min 1st harmonics = 2.4 Hz 2nd harmonics = 3.6 Hz Frequency (Hz) DC noise, baseline movement Controlled respiration, 0.25 Hz Frequency (Hz) Fig. 2. Frequency analysis result of PPG waveform (subject 11). (a) Fast Fourier transform result of PPG during 0.25 Hz respiration (subject s average heart rate was 71.6 bpm). PPG is composed with major frequency component around 1.2 Hz and its harmonics. (b) Magnification of dashed box of (a). Respiration effect and baseline movement (o0.1 Hz) could be found to remove.

5 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) time (s) Fig. 3. An example of PPG signal conditioning and peak detection result by adaptive threshold method (subject 11). (a) Original PPG waveform which contains respiratory effect and baseline wandering. Respiration is controlled with 0.25 Hz. (b) Hz bandpass filtered PPG waveform. (c) V max peak detection result. (d) V min peak detection result. Dashed line is detection threshold, and circle means detected peaks Peak detection and evaluation An example of peak detection results using ADT method are shown in Fig. 3(c) and (d). V max and V min were detected in Fig. 3(c) and (d), respectively. Dashed line represents detection threshold and circle means V max and V min. Tables 2 and 3, respectively, show the statistical result and comparison for the V max and V min detection by LCM method and ADT method. In V max detection, average SE was 82.47% and 98.22% for a LCM and ADT, respectively. Average FD was 17.53% and 1.78% in LCM and ADT, and +P was 100% in every method. In V min detection, SE was 87.10% and 98.87%, +P was % and 99.08% and FD was 12.90% and 1.14% in a LCM and ADT, respectively. 4. Discussion 4.1. Detection accuracy Results from the present study highlight the detection accuracy of the adaptive threshold detection method for PPG V max and V min detection. It is appeared that proposed signal conditioning and peak detection algorithm have more than 98% SE and 99% +P in both V max and V min detection. Failed detection rate was o2% for both features. ADT shows predominating detection performance compared with LCM. Average SE was improved in both V max (15.10%) and V min (11.42%) detections. Moreover, FDR was decreased by 15.10% and 11.40% for V max and V min detection, respectively. From the V max detection result of protocols 2 and 4, it was appeared that detection performances are significantly decreased in LCM method. SE was decreased by 12.62% and 9.17% for protocols 2 and 4. This means that LCM method has the difficulties such as slow respiration (protocol 2) and physiological variation from posture changing (protocol 4) on V max detection. ADT method showed uniform detection performance, and it implies ADT is robust method to respiration effect and physiological variation. For a reliable detection, detection results also could be considered with fluctuations of detection rate. Standard deviation of LCM method sensitivity was 4.98% and 2.15% for V max and V min detection. However, in ADT method, 0.88% and 0.14% were reported as a standard deviation of sensitivity. From these result, it is suggested that proposed ADT method has a better detection performance and detection stability than previous LCM method in V max and V min detection of PPG.

6 1150 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) Table 2 Peak detection result for the V max detection. Protocol Position Respiration Method V max Ref. a (beats) TP (beats) FP (beats) FN (beats) FD (beats) SE b (%) +P c (%) FDR (%) 1 Supine Spon d LCM e ADT f (667) ( ) ( 667) ( 667) (10.82) ( ) ( 10.82) 2 Supine 0.1 Hz LCM ADT (1261) ( ) ( 1261) ( 1261) (22.95) ( ) ( 22.95) 3 Supine 0.25 Hz LCM ADT (563) ( ) ( 563) ( 563) (9.75) ( ) ( 9.75) 4 Sitting Spon LCM ADT (924) ( ) ( 924) ( 924) (17.82) ( ) ( 17.82) Total LCM ADT (3415) ( ) ( 3415) ( 3415) (15.10) ( ) ( 15.10) Parenthesis represents the fluctuation between LCM and ADT method results. a Number of reference peaks. b Sensitivity. c Positive predictivity. d Spontaneous respiration. e Local minima/maxima. f Adaptive threshold. Table 3 Peak detection result for the V min detection. Protocol Position Respiration Method V min Ref. a (beats) TP (beats) FP (beats) FN (beats) FD (beats) SE b (%) +P c (%) FDR (%) 1 Supine Spon d LCM e ADT f ( ) (671) (1) ( 671) ( 670) (10.89) ( 0.02) ( 10.87) 2 Supine 0.1 Hz LCM ADT ( ) (592) (1) ( 592) ( 591) (10.77) ( 0.02) ( 10.76) 3 Supine 0.25 Hz LCM ADT ( ) (537) (0) ( 537) ( 537) (9.29) ( ) ( 9.29) 4 Sitting Spon LCM ADT ( ) (784) (3) ( 781) ( 781) (15.12) ( 0.06) ( 15.06) Total LCM ADT ( ) (2584) (5) ( 2584) ( 2579) (11.42) ( 0.02) ( 11.40) Parenthesis represents the fluctuation between LCM and ADT method results. a Number of reference peaks. b Sensitivity. c Positive predictivity. d Spontaneous respiration. e Local minima/maxima. f Adaptive threshold Respiration effect and physiological investigation It is postulated that blood flow and heart rate are affected by physical and physiological characteristics of respiration by assuming that the major reason of miss-detection is blood flow disturbance such as low perfusion. It was demonstrated that the slower respiratory rates, the higher tidal and the smaller minute volumes are present [26]. This result is consistent with our study, suggesting that respiration could change PPG morphology and it disturbs peak detection. However, it should be considered that there are the difference between V max and V min detection result. Respiration greatly affects on LCM detection results. In controlled respiration, V max and V min detection rate had a different tendency, SE of V max was decreased by 12.62% in 0.1 Hz respiration, but SE of V min was not significantly changed. These result supports that respiration control could be a reason for a physiological variation such as heart rate, stroke volume (SV) and total peripheral resistance (TPR), and these parameters are an

7 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) important component to decide PPG waveform. Especially, considering that V max detection is closely related on reflected wave, it is confirmed that respiration could change physiological status, and it also have an effect on V max detection of PPG Limitations Though we revealed over than 98% SE and +P in this study, there is an important limitation. Differently from V min, V max could be severely affected from reflected wave. Generally, PPG is described with two neighbor peaks and preceding peak (incident wave) has higher amplitude than later peak (reflected wave). However, reflected wave could be large, and the distance between incident wave and reflected wave becomes more closely by aging, hypertension and arterial sclerosis. From this phenomenon, incident and reflected wave boundary becomes ambiguously, even they are merged [33]. In this case, it would be hard to classify incident wave peak by proposed peak detection method. This is a common limitation of PPG peak detection algorithm based on maxima value, and it could be a reason for a miss detection and timing error. In presented study, all of subjects are young and healthy, and nobody has a reported cardiovascular disease therefore there is no classification problem related on wave merging. Another major issue mentioned and unsolved by this study is the low perfusion problem. Low perfusion is a different issue from the peak detection, it is rather closely related on signal measurement or conditioning. Therefore, low perfusion should be dealt from signal acquisition schemes such as sensor design to advanced signal processing in further study. 5. Conclusion From this research, it is concluded that frequency characteristic of PPG is correspond with peripheral blood flow characteristic defined by previous research based on frequency characteristic of arterial pressure [31]. Proposed adaptive threshold detection method shows slightly different detection performance in situations. However, in static environment, peak detection results are confirmed over 98% SE and +P. Furthermore, it is suggested very useful method in real-time peak detection because PPG is frequently used for the real-time application. Though physiological studies and individual adaptive controlling method are needed to further improvement for robust detection, it is our belief that a proposed method can provide for a simple and reliable PPG peak detection. 6. Summary Photoplethysmography (PPG)-based temporal analyses have been widely used as a useful analytical method in physiological and cardiovascular diagnosis. Most of temporal approaches of PPG are based on detected peak points, peak and foot of PPG. The aim of presented study is the development of improved peak detection algorithm of PPG waveform. Our study suggests the frequency filtering method based on PPG frequency characteristic, and adaptive threshold method (ADT) for PPG peak detection. The performance of the proposed method was evaluated on 18 young and healthy subjects in the supine position. To verify the reliability of ADT, experiment was performed with various respiration rate and posture changes. Respiration was controlled by 0.1 and 0.25 Hz to induce a respiratory distortion on PPG, and spontaneous respiration was used as a reference. For comparative study, PPG was also measured in the sitting position. Detection accuracy was compared with traditional peak detection method, local minima/maxima detection method (LCM). From the investigation of frequency characteristic, it was confirmed that baseline drift and respiration effect were concentrated in low frequency range (o0.5 Hz) and it could be removed by bandpass filtering; however, low perfusion problem was still remains. Overall, with 22,623 pulses, the sensitivity, positive predictivity and failed detection rate of the ADT was 98.04%, % and 1.96% for V max, and 98.84%, 99.98% and 1.18% for V min, respectively. Moreover, failed detection rate was decreased by 15.10% and 11.40% in V max and V min detection compared with LCM. The present study demonstrates a promising approach to overcome respiration effect and to detect PPG peak. More extensive investigation is necessary to adapt for the cardiovascular diseases, whose PPG morphology has different form. Conflict of interest statement None to declare. References [1] J. Allen, Photoplethysmography and its application in clinical physiological measurement, Physiol. Meas. 28 (2007) R1 R39. [2] J. Allen, A. Murray, Similarity in bilateral photoplethysmographic peripheral pulse wave characteristics at the ears, thumbs and toes, Physiol. Meas. 21 (2000) [3] J. Allen, A. Murray, Variability of photoplethysmography peripheral pulse measurementsat the ears, thumbs and toes, IEE Proc. Sci. Meas. Technol. 147 (2000) [4] M. Nitzan, B. Khanokh, Y. Slovik, The difference in pulse transit time to the toe and finger measured by photoplethysmography, Physiol. Meas. 23 (2002) [5] R. Asmar, A. Benetos, J. Topouchian, P. Laurent, B. Pannier, A.M. Brisac, R. Target, B.I. Levy, Assessment of arterial distensibility by automatic pulse wave velocity measurement. Validation and clinical application studies, Hypertension 26 (1995) [6] J.R. Jago, A. Murray, Repeatability of peripheral pulse measurements on ears, fingers and toes using photoelectric plethysmography, Clin. Phys. Physiol. Meas. 9 (1988) [7] G.S. Chan, P.M. Middleton, B.G. Celler, L. Wang, N.H. Lovell, Automatic detection of left ventricular ejection time from a finger photoplethysmographic pulse oximetry waveform: comparison with Doppler aortic measurement, Physiol. Meas. 28 (2007) [8] J.Y. Foo, C.S. Lim, P. Wang, Evaluation of blood pressure changes using vascular transit time, Physiol. Meas. 27 (2006) [9] R.A. Payne, C.N. Symeonides, D.J. Webb, S.R. Maxwell, Pulse transit time measured from the ECG: an unreliable marker of beat-to-beat blood pressure, J. Appl. Physiol. 100 (2006) [10] A. Benetos, K. Okuda, M. Lajemi, M. Kimura, F. Thomas, J. Skurnick, C. Labat, K. Bean, A. Aviv, Telomere length as an indicator of biological aging: the gender effect and relation with pulse pressure and pulse wave velocity, Hypertension 37 (2001) [11] A. Porta, C. Gasperi, G. Nollo, D. Lucini, P. Pizzinelli, R. Antolini, M. Pagani, Global versus local linear beat-to-beat analysis of the relationship between arterial pressure and pulse transit time during dynamic exercise, Med. Biol. Eng. Comput. 44 (2006) [12] M. Nitzan, I. Faib, H. Friedman, Respiration-induced changes in tissue blood volume distal to occluded artery, measured by photoplethysmography, J. Biomed. Opt. 11 (2006) [13] F. Yasuma, J. Hayano, Respiratory sinus arrhythmia: why does the heartbeat synchronize with respiratory rhythm?, Chest 125 (2004) [14] R. Pallas-Areny, J. Colominas-Balague, F. Rosell, The effect of respirationinduced heart movements on the ECG, IEEE Trans. Biomed. Eng. 36 (1989) [15] L. Xu, D. Zhang, K. Wang, N. Li, X. Wang, Baseline wander correction in pulse waveforms using wavelet-based cascaded adaptive filter, Comput. Biol. Med. 37 (2007) [16] S. Lu, H. Zhao, K. Ju, K. Shin, M. Lee, K. Shelley, K.H. Chon, Can photoplethysmography variability serve as an alternative approach to obtain heart rate variability information?, J. Clin. Monit. Comput. 22 (2008) [17] J.I. Davies, A.D. Struthers, Pulse wave analysis and pulse wave velocity: a critical review of their strengths and weaknesses, J. Hypertens. 21 (2003) [18] G.M. London, A. Guerin, Influence of arterial pulse and reflective waves on systolic blood pressure and cardiac function, J. Hypertens. Suppl. 17 (1999) S3 S6. [19] G.M. London, A.P. Guerin, Influence of arterial pulse and reflected waves on blood pressure and cardiac function, Am. Heart J. 138 (1999)

8 1152 H.S. Shin et al. / Computers in Biology and Medicine 39 (2009) [20] W.W. Nichols, D.G. Edwards, Arterial elastance and wave reflection augmentation of systolic blood pressure: deleterious effects and implications for therapy, J. Cardiovasc. Pharmacol. Ther. 6 (2001) [21] M. O Rourke, Arterial compliance and wave reflection, Arch Mal Coeur Vaiss 84 Spec No 3 (1991) [22] M.F. O Rourke, Isolated systolic hypertension, pulse pressure, and arterial stiffness as risk factors for cardiovascular disease, Curr. Hypertens. Rep. 1 (1999) [23] M.F. O Rourke, D.E. Gallagher, Pulse wave analysis, J. Hypertens. Suppl. 14 (1996) S147 S157. [24] M.F. O Rourke, R.P. Kelly, Wave reflection in the systemic circulation and its implications in ventricular function, J. Hypertens. 11 (1993) [25] M.F. O Rourke, T. Yaginuma, Wave reflections and the arterial pulse, Arch. Intern. Med. 144 (1984) [26] R. Baron, H.J. Habler, K. Heckmann, H. Porschke, Respiratory modulation of blood flow in normal and sympathectomized skin in humans, J. Auton. Nerv. Syst. 60 (1996) [27] N. Westerhof, M.F. O Rourke, Haemodynamic basis for the development of left ventricular failure in systolic hypertension and for its logical therapy, J. Hypertens. 13 (1995) [28] J. Allen, A. Murray, Age-related changes in peripheral pulse timing characteristics at the ears, fingers and toes, J. Hum. Hypertens. 16 (2002) [29] J. Allen, A. Murray, Age-related changes in the characteristics of the photoplethysmographic pulse shape at various body sites, Physiol. Meas. 24 (2003) [30] K. Nakajima, T. Tamura, H. Miike, Monitoring of heart and respiratory rates by photoplethysmography using a digital filtering technique, Med. Eng. Phys. 18 (1996) [31] M. O Rourke, M. Taylor, Vascular impedance of the femoral bed, Circ. Res. 18 (1966) [32] H. Shin, C. Lee, M. Lee, Ideal filtering approach on DCT domain for biomedical signals: index blocked DCT filtering method (IB-DCTFM), J. Med. Syst Available online at / u0224h7m6u/S. [33] W.W. Nichols, M.F. O Rourke, McDonald s Blood Flow in Arteries, Hodder Arnold, 2004 pp Hang Sik Shin received MS. degree in electrical and electronic engineering from the Yonsei University, Seoul, Republic of Korea in He is now pursuing the Ph.D. degree in biomedical engineering from the same university. His research interests focus on biomedical signal processing and non-invasive physiological measurement and applications. Chungkeun Lee received MS. degree in electrical and electronic engineering from the Yonsei University, Seoul, Republic of Korea in He is now pursuing the Ph.D. degree in electrical and electronic engineering from the same university. His research interests focus on cardiovascular-respiratory coupling analysis and applications. Myoungho Lee received MS. degree and Ph.D. degree in biomedical engineering from the Yonsei University, Seoul, Republic of Korea in 1974 and 1978, respectively. He is a chair of IFMBE oriental and alternative medicine working group, and a chief of Korea e-health research center and Korea e-health association. Since 1980, he has been with the Yonsei university, where is a professor of electrical and electronics engineering. His research interests focus on biomedical applications and e-health system.