Continuous Cuffless Blood Pressure Estimation Using ...

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TBME.2015.2480679, IEEE Transactions on Biomedical Engineering

Continuous Cuffless Blood Pressure Estimation Using Pulse Transit Time and Photoplethysmogram Intensity Ratio Xiao-Rong Ding, Student Member, IEEE, Yuan-Ting Zhang, Fellow, IEEE, Jing Liu, Wen-Xuan Dai, Hon Ki Tsang*, Senior Member, IEEE  Abstract—Pulse transit time (PTT) has attracted much interest for cuffless blood pressure (BP) measurement. However, its limited accuracy is one of the main problems preventing its widespread acceptance. Arterial BP oscillates mainly at high frequency (HF) because of respiratory activity, and at low frequency (LF) because of vasomotor tone. Prior studies suggested that PTT can track BP variation in HF range, but was inadequate to follow the LF variation, which is probably the main reason for its unsatisfactory accuracy. This paper presents a new indicator, the photoplethysmogram intensity ratio (PIR) which can be affected by changes in the arterial diameter and thus trace the LF variation of BP. Spectral analysis of BP, PTT, PIR and respiratory signal confirmed that PTT was related to BP in HF at the respiratory frequency, while PIR was associated with BP in LF range. We therefore develop a novel BP estimation algorithm by using both PTT and PIR. The proposed algorithm was validated on 27 healthy subjects with continuous Finapres BP as reference. The results showed that the mean ± standard deviation (SD) for the estimated systolic, diastolic and mean BP with the proposed method against reference were -0.37±5.21 mmHg, -0.08±4.06 mmHg, -0.18±4.13 mmHg, and mean absolute difference (MAD) were 4.09 mmHg, 3.18 mmHg, 3.18 mmHg, respectively. Furthermore, the proposed method outperformed the two most cited PTT algorithms for about 2 mmHg in SD and MAD. These results demonstrated that the proposed BP model using PIR and PTT can estimate continuous BP with improved accuracy. Index Terms—Arterial diameter change, cuffless blood pressure, photoplethysmogram intensity ratio, pulse transit time, respiration, vasomotion

This work was supported in part by the Guangdong Innovation Research Team Fund for Low-cost Healthcare Technologies in China, the External Cooperation Program of the Chinese Academy of Sciences (Grant GJHZ1212). Asterisk indicates corresponding author. Xiao-Rong Ding, Yuan-Ting Zhang, Jing Liu, and Wen-Xuan Dai is with the Joint Research Centre for Biomedical Engineering, Department of Electronic Engineering, The Chinese University of Hong Kong, Hong Kong SAR, China (e-mail: [email protected]; [email protected]; [email protected]; [email protected]) *Hon Ki Tsang is with the Department of Electronic Engineering, The Chinese University of Hong Kong, Hong Kong SAR, China (e-mail: [email protected]). Copyright (c) 2015 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending an email to [email protected].

I. INTRODUCTION

B

pressure (BP) is an important hemodynamic parameter varying between systolic BP (SBP) to diastolic BP (DBP) in each heartbeat. High BP, also known as hypertension, is one of the major modifiable risk factors leading to the development of cardiovascular diseases (CVDs) – the number one killer in the world. Hypertension is highly prevalent but poorly controlled because of the low awareness and treatment rate [1], which enhances development of CVDs and results in significant burdens on individuals and society. BP variability (BPV) has been reported to have prognostic value for hypertension [2], and thus continuous BP measurement is crucial for early prevention, detection, evaluation and treatment of hypertension and related CVDs. Conventional 24-hour ambulatory BP monitoring can facilitate monitoring of BPV through measuring BP at regular intervals with auscultatory or oscillometric approaches. However, it has limitations including the discontinuous nature and the discomfort caused by the repeated cuff inflations. Compared with cuff-based BP techniques, pulse transit time (PTT) method has received much attention over the recent decades because of its capability to track BP change, as well as its advantages as a noninvasive, continuous and most importantly cuffless tool for BP measurement [3-5]. PTT is the time taken by the arterial pulse propagating from the heart to a peripheral site, and can be calculated as the time interval between the R wave peak of electrocardiogram (ECG) and a characteristic point of photoplethysmogram (PPG). The fundamental principle of PTT-based method is based upon the pulse wave velocity (PWV) recording through the Moens-Korteweg (M-K) equation: LOOD

PWV 

Eh d

(1)

which relates PWV with the elastic modulus of vessel wall E, blood density ρ and arterial dimension properties such as vessel thickness h and arterial diameter d. PWV is inversely related with PTT, i.e., PWV=K/PTT, where K is the distance between heart and certain peripheral site; and E can be exponentially correlated BP through the following equation [6]:

0018-9294 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.


E  E0e

P

(2)

where E0 is the elastic modulus at zero pressure; γ is a coefficient depending on particular vessel, and is BP. Therefore, PTT can be translated into BP with an initial calibration under the assumption that h/d keeps constant. PTT-based BP estimation has been extensively studied ever since 2000 [7-20], when Chen et al [7] estimated SBP using pulse arrival time with intermittent calibration, and showed that the estimated SBP was highly correlated with reference SBP (r=0.97±0.02). In 2005, Poon et al [9] established a model with PTT to estimate BP with initial calibration, and achieved an accuracy of 0.6±9.8 mmHg and 0.9±5.6 mmHg for SBP and DBP, respectively. Recently, Wibmer et al [19] investigated the relationship of PTT and SBP through regression analysis and found a nonlinear approach was better than linear one. Although PTT has been considered a promising surrogate of BP and could become the most widely used technique for noninvasive continuous BP monitoring in the future [4, 5], there are still several problems to be solved before its widespread application. Frist, some PTT-BP models could only provide one BP parameter, e.g., exclusively SBP [7, 8, 21, 22], DBP [11], or mean BP (MBP) [18], but SBP, DBP, and MBP all have clinical significance. Second, a calibration procedure is required to map PTT to BP. However, re-calibration at intermittent intervals is often necessary for accurate estimation, potentially owing to the inadequacy of PTT to track BP variation over a long period. Last and most importantly, the accuracy of PTT-based BP estimation is unsatisfactory. The possible reasons are the influences of the vascular or vasomotor tone and the pre-ejection period (PEP) [3]. Regarding PEP issue, impedance cardiogram (ICG), phonocardiogram (PCG) [20], ballistocardiogram (BCG) or two peripheral PPG have been adopted to eliminate the effect of PEP. Nonetheless, several studies indicated that PTT with PEP included actually performs better for BP estimation than that with PEP removed, which demonstrates the positive effect of PEP on BP estimation [8, 22, 23]. For the vasomotor tone, previous research has examined its influence on BP-PTT relationship [24], and central PPG instead of peripheral PPG was suggested to alleviate such effect [18]. However, few studies have attempted to take this factor into account in the PTT-BP estimation model to improve the accuracy. BP is dynamic and its rhythmic oscillations can be identified with the appearance in its spectrum as individual peaks, which reflect: (1) the oscillations with a frequency typically between 0.2-0.35 Hz, a frequency similar to that of normal respiratory activity, defined as high frequency (HF); (2) oscillations with a frequency of approximately 0.1-0.15 Hz, suggesting the sympathetic modulation of vasomotor tone, defined as low frequency (LF) [25-27]. According to our prior research work [28, 29], PTT could track BP in HF range, but was inadequate to follow LF variations in BP. This is probably the most important reason for the inaccuracy of estimated BP with only

the PTT and the requirement of intermittent calibration to maintain accuracy. However, to date, few study has addressed the LF component in cuffless BP estimation. Here we propose a new indicator, the PPG intensity ratio (PIR) [30], that can reflect changes in the arterial diameter and thereby the arterial vasomotion, and thus allow tracking of BP in the LF range. Furthermore, we develop a novel BP estimation algorithm which employs both the PIR and the PTT to improve the estimation accuracy. II. METHODOLOGY A. PTT and PIR Fig. 1 illustrates the diagram of PTT and PIR calculation, where PTT is determined as the time interval between the R wave peak of ECG and the peak of first derivative of PPG in the same cardiac cycle, and PIR is the ratio of PPG peak intensity IH to PPG valley intensity IL of one cardiac cycle. Our earlier study shows that PIR can theoretically reflect the arterial diameter change △d during one cardiac cycle from systole to diastole, and PIR is exponentially linked with △d through the following expression [30]:

PIR  e d

(3)

where α is considered to be a constant related to the optical absorption coefficients in the light path.

Fig. 1. Diagram of pulse transit time (PTT) and photoplethysmogram (PPG) intensity ratio (PIR) calculation, where IH indicates PPG peak intensity, IL the valley intensity, and 1st dPPG is the first derivative of PPG.

From a physiological perspective, BP is mainly affected by four factors: arterial compliance, cardiac output, peripheral resistance, and blood volume [31]. Arterial compliance can be evaluated by PTT, since PTT is an index of arterial stiffness [32]. Also, cardiac output can be related to PTT through heart rate. With regard to peripheral resistance and blood volume, one of the primary sources is the arterial diameter change which can be assessed by PIR as described above. Accordingly, PTT and PIR can capture BP variations indirectly, and be used for BP estimation.



B. Model-based BP Estimation with PTT and PIR Arterial BP is a hemodynamic parameter that fluctuates on a beat-to-beat basis as a result of the dynamic interplay involving vasomotion, arterial mechanisms and neural regulation [33]. Beat-to-beat BP fluctuations are usually attributed to two rhythmic events: respiration and vasomotion [34-36]. The respiratory rhythm is a HF spectral component that occurs in BP variability, and also considered to be a marker of vagal modulation, whereas the slow oscillation, corresponding to the vasomotor waves, is a LF component that is present in BPV, and is also a marker of sympathetic modulation. Physiological study of BPV showed that both the slow variability and fast variability can be observed in SBP, while the DBP exhibits only on the slow variability [37]. Since SBP is the summation of pulse pressure (PP) and DBP, it is hypothesized that the HF component is mainly dominant in PP. Furthermore, PTT and PIR have been investigated to reflect HF and LF component in BPV [30], respectively. It is therefore speculated that PP and DBP can be derived from PTT and PIR, respectively, and consequently the SBP can be estimated accurately. 1) PP Estimation with PTT Based on M-K Equation In the M-K equation (1), the elastic modulus E is given by [38]:





P 2 1   2 RO Ri E  2 2 RO RO  Ri

2

1 PTT

(6)

With initial calibrated PP0 and PTT0, PP can be derived in terms of measured PTT:  PTT0  PP  PP0     PTT 

(8)

where P0 is the end-systolic aortic pressure. Since C is constant in a relatively short period, DBP mainly varies with R. Noting that the major regulator of peripheral vascular resistance is the vessel diameter, R will mainly rely on the arterial diameter change △d. As mentioned above in (3), △d is related with PIR as follows: 1  e d PIR

(9)

And △d is inversely related to R. Thus DBP depends on the reciprocal of PIR: 1 PIR

(10)

Therefore, DBP can be derived with calibrated DBP 0 and PIR0.

(5)

1 PTT 2

DBP  P0  et / RC

(4)

Since the artery radius change is quite small compared with elastic modulus change, it is assumed to be constant. According to (1) and (4)-(5), △P has a relationship with PTT as follows: PP 

In a “pure” Windkessel, the DBP can be theoretically expressed in terms of RC using the following equation:

DBP 

where RO is the external radius, △RO is the external radius change in response of the pressure change △P, and △P is the PP in the artery; Ri is the internal radius, and σ is the Poisson’s ratio. PWV is reversely proportional to PTT: PWV 

Fig. 2. The two-element Windkessel [39].

2

(7)

2) DBP Estimation with PIR based on Windkessel Model Two-element Windkessel model, originally proposed by Frank, consists of peripheral resistance R and arterial compliance C [39], as shown in Fig. 2.

DBP  DBP0 

PIR0 PIR

(11)

3) SBP Estimation with PP and DBP SBP is the sum of PP and DBP. Thereupon, beat-to-beat SBP can be estimated with the addition of (7) and (11): SBP  DBP0 

PIR0  PTT0   PP0    PIR  PTT 

2

(12)

C. Experiment To validate the proposed BP estimation using both PTT and PIR as given by (11) and (12), an experiment was conducted on 27 healthy adults (14 males) with mean age of 25.6±2.1 years (range 21-29 years), who were nonsmokers with no history of cardiovascular disease. Reference BP was measured by Finapres (Finapres Medical System), a noninvasive continuous BP measurement system, with the finger cuff on the left thumb, and brachial cuff on the left upper arm. ECG and PPG were acquired with one-lead ECG electrodes placed on left and right arms, and PPG sensor on left index finger, respectively. Synchronous respiratory activity continuously monitored by recording the chest movement with respiratory monitoring belt (Vernier Software & Technology). All tests were performed with subjects in the seated position, and the signals were



recorded at the sampling rate of 1000 Hz for five minutes. All the subjects gave their informed consent prior to the experiments, in accordance with the guidelines of the Institutional Research Ethics Board. D. Signal Processing and Data Analysis In order to verify the capability of PTT to track the HF component of SBP as well as respiratory activity, and PIR to reflect LF fluctuations of BP, the power spectrum analysis of SBP, DBP, PP, respiratory signal, PTT and PIR were conducted in 0-0.5 Hz frequency range based on Lomb-Scargle periodogram method [40]. Difference mean and standard deviation (SD), as well as mean absolute difference (MAD) between estimated BP with the proposed method and reference BP were used as the evaluation metrics. The agreement between reference BP and estimated BP with the proposed method were analyzed according to the Bland-Altman approach [41], with the agreement limits defined by mean ±1.96×SD. In addition, the proposed method was compared with two most cited PTT-based algorithms [7, 9] for cuffless BP estimation. Statistical significance was estimated using Student’s t-test. P