This content has been downloaded from IOPscience. Please scroll down to see the full text.

Download details:

IP Address: 131.155.2.68

This content was downloaded on 22/02/2017 at 08:19

Please note that terms and conditions apply.

Arterial path selection to measure pulse wave velocity as a surrogate marker of blood

pressure

View the table of contents for this issue, or go to the journal homepage for more

2017 Biomed. Phys. Eng. Express 3 015022

(http://iopscience.iop.org/2057-1976/3/1/015022)

Home Search Collections Journals About Contact us My IOPscience

You may also be interested in:

A survey on signals and systems in ambulatory blood pressure monitoring using pulse transit time

Dilpreet Buxi, Jean-Michel Redouté and Mehmet Rasit Yuce

Cuff-less blood pressure measurement using pulse arrival time and a Kalman filter

Qiang Zhang, Xianxiang Chen, Zhen Fang et al.

Impact of heart disease and calibration interval on accuracy of pulse transit time–based blood

pressure estimation

Xiaorong Ding, Yuanting Zhang and Hon Ki Tsang

Non-constrained monitoring of systolic blood pressure on a weighing scale

Jae Hyuk Shin, Kang Moo Lee and Kwang Suk Park

A new approach for non-intrusive monitoring of blood pressure on a toilet seat

Jung Soo Kim, Young Joon Chee, Ju Wan Park et al.

Difference in pulse transit time to the toe and finger

M Nitzan, B Khanokh and Y Slovik

Effect of confounding factors on BP estimation using PAT

Jung Soo Kim, Ko Keun Kim, Hyun Jae Baek et al.

Non-invasive monitoring of pulmonary artery pressure from timing information by EIT: experimental

evaluation during induced hypoxia

Martin Proença, Fabian Braun, Josep Solà et al.

Biomed. Phys. Eng. Express 3 (2017) 015022 https://doi.org/10.1088/2057-1976/aa5b40

PAPER

Arterial path selection tomeasure pulse wave velocity as a surrogatemarker of blood pressure

MRadha1,2, G Zhang1,2, J Gelissen1, K deGroot1, RHaakma1 andRMAarts1,2

1 PersonalHealth, Philips Research, Royal PhilipsHighTechCampus 34, 5656AE, Eindhoven, TheNetherlands2 Electrical Engineering, EindhovenUniversity of Technology Flux,floor 7, POBox 513, 5600 MB, Eindhoven, TheNetherlands

E-mail:mustafa.radha@philips.com

Keywords: photoplethysmography, pulse arrival time, systolic blood pressure, sensor locations, hemodynamics, pulse wave velocity, pulsepropagation

AbstractThe velocity of the propagating arterial pulsewave (pulse wave velocity, PWV)has been proposed asan unobtrusive and possibly continuous surrogatemeasure of systolic blood pressure (SBP). PWV isderived from the arrival time of the blood pulse at a peripheral arterial location,most often the finger.Reported performances were not yet accurate enough for clinical application but good enough as anunobtrusive surrogate in other settings. However, the finger PPG is not an ideal location in the homesetting as it obstructs handmovement and can suffer fromperipheral vasomotion and orthostaticpressure changes. In this paper we examine the viability of other pulse arrival locations for themeasurement of PWV. PWVwas derived to the finger (most common location), wrist (less obtrusivelocation), ear (more proximal) and ankle (more distal). Correlation analysis for PWV from eachlocationwith SBPwas performed and the calibration procedure was studied.Wrist PWVaccuracy isfound to be comparable tofinger PWV in terms of correlation and estimation error with SBP. The earPWV, being a theoretically favorable location, is shown to have a larger inter-subject variance in thecalibration procedure compared to other locations. Ankle PWV shows stable calibration parametersacross subjects but Bland-Altmann analysis reveals unusual error trends. In conclusion, while resultsindicate that all sensor locations are usable to some extent, there are still some distinct propertiesassociatedwith each sensor location that should be taken into accountwhen designing an SBPalgorithmbased onPWV.

1. Introduction

Blood pressure (BP) is one of the major vital signs anda strong indicator of cardiovascular health. Theprevalence of pathologically high BP (i.e. hyperten-sion) is higher than ever (Cutler et al 2008) and newstudies reveal that even moderate elevations in bloodpressure that were not considered pathological in thepast are also linked to cardiovascular events (SPRINTResearch Group 2015). These trends are met with anincreasing demand for better blood pressure manage-ment, of which an essential ingredient is the measure-ment of BP. Next to the clinical use of BPmeasurement for hypertension diagnosis and otherhaemodynamic complications (Chase et al 2004,Greenland et al 2010), home blood pressure monitor-ing is also important (ESH/ESC 2013). Currently,

individuals are capable of self-monitoring BP at homeby using a standard oscillometric pressure cuff. Thistype of devices calculate the systolic anddiastolic bloodpressure (SBP and DBP) which correspond respec-tively to the peak pressure in the artery as the pulsewave is passing through and the lowest pressure in-between beats, two of the most recognized BPparameters. However, these pressure cuffs lack port-ability and their inflation is obtrusive and painful.These factors make the measurement of BP a burdenon the individual, decreasing the tendency of users toroutinely self-monitor BP and thus negatively con-tribute to blood pressuremanagement. This motivatesresearch into new methods of surrogate BP measure-ment. Besides the cuff, a few solutions have beenproposed over the past decades (Pickering et al 2005).However, these productswere not designedwith home

RECEIVED

7December 2016

REVISED

20 January 2017

ACCEPTED FOR PUBLICATION

23 January 2017

PUBLISHED

8 February 2017

© 2017 IOPPublishing Ltd

measurement in mind: while they enable long-termcontinuous measurement, they are still bulky andobtrusive (e.g. vascular unloading technique) or diffi-cult to operate (e.g. applanation tonometry). For anoverview of current BP measurement methods, thereader is referred to (Pickering et al 2005).

1.1. Pulsewave velocityOne of the most promising surrogate methods for BPmeasurement is pulse arrival time (PAT), which is thetime the pulse wave takes to arrive at an arterial siteafter ejection from the heart. PAT is an indicator ofpulse wave velocity (PWV), which can be derived bydividing the length L of the traversed arterial segmentover PAT:

( )=L

PWVPAT

. 1

The onset of ventricular ejection is most oftenmeasured with electrocardiogram (ECG) even thoughit is known that there is a slight time differencebetween the ECG’s R-peak and the actual start of pulsetransit, known as the pre-ejection period (Payneet al 2006, Zhang et al 2011). Other more accuratepulse onset measurement methods exist but are eithermore difficult to operate (e.g. phonocardiogram),noise-prone (e.g. seismocardiogram and ballisto-cardiogram Inan et al 2015) or expensive (e.g. ultra-sound). Pulse arrival is conveniently measured withphotoplethysmography (PPG), an easy to use andaffordable sensor. PPG works by illuminating the skinwith light and capturing the intensity of the reflectedlight from the skin. The intensity of reflection dependson how much of the light is absorbed by the skin. Bychoosing a light frequency that is absorbed by blood, itis possible to measure the pulsatile blood flow. Multi-ple morphological markers on the PPG pulse can beused for the detection of pulse arrival (Kortekaaset al 2012). An important mediator in the reflectionprofile is the contact pressure of the PPG sensor: aloose contact pressure could allow ambient light to bepicked up by the photodetector while a high pressurecan decrease blood perfusion locally, both reducingthe quality of the signal (Teng andZhang 2007).

1.2. The PWV-BP relationPWV is classically related to BP through the Moens-Korteweg equations (Proença et al 2010). The relation-ship between BP and PWV can be derived from thisequation as follows:

··

( )r

=gE e h

rPWV

2, 20

BP

where E0 is the elastic modulus at zero pressure, ρ isthe blood density, h is the vessel radius, r is the vesselwall thickness and γ is a coefficient depending on theparticular vessel. This function can be rewritten as:

( ) ( ) ( )r

g= +E h

rln PWV

ln

2BP 32 0

which can be simplified as:

( ) ( ) ( )g r g

= -E h

rBP

2ln PWV

ln

2. 40

Thus, when assuming that h, r, γ and and ρ staymore or less constant from measurement to measure-ment (in comparison with PWV), it can be proposedthat there is a linear dependence between BP andln(PWV) (Peter et al 2014, Mukkamala et al 2015).Thus, a calibration for each arterial segment (and eachperson) looks as follows: · ( )= +a bBP ln PWV ,

where =g

a 2and ( )= -

r gb E h

r

ln

20 . While many encoura-

ging results have been achieved with this way of work-ing, it has also been evident that the approach has itslimitations. For clinical use, this theoretical model didnot prove to be accurate enough to deal with the com-plex and abnormal physiologies of patients (Smithet al 1999) and it has been found to be an inaccuratemarker for both diastolic and mean arterial pressure(Payne et al 2006). However, for SBP many positiveresults have been obtained in various contexts (Solàet al 2009, Gesche et al 2012, Tang et al 2016).

Recently, wearable technology for healthmonitor-ing has enjoyed a wide use and is becoming part ofeveryday life. This has enabled sophisticated healthmonitoring methods to become accessible to many.Such technology that is not primarily intended to diag-nose or treat disease falls under the category of generalwellness technology. These devices do not have tomeetthe rigorous standards of medical equipment whilestill can play a supplementary role in maintaining ahealthy lifestyle (U.S. Food and Drug Administra-tion 2015). PWV as a surrogate SBP measurementmethod could be positioned within this landscape.However, there are still practical issues that need to beresolved before PWV-based SBP measurement can beembodied within an application that is readily deploy-able. While the theoretical model from equation (4)holds in controlled conditions, in a real embodimentthere are a number of practical issues that can heavilyimpact accuracy.

1.3. Practical issuesThe most common pulse arrival location in literatureused to study PWV as a surrogate marker of SBP is onthe finger (Wong et al 2009, Peter et al 2014, Mukka-mala et al 2015, Sun et al 2016b), mainly because of thehistorical reason that the finger PPG clip has been anintegral part of clinical practice (used for the assess-ment of blood oxygen saturation), making themmoreaccessible for clinical studies (Yoon et al 2002,Wijshoffet al 2016). While this has enabled fundamentalresearch on the topic, it is not the most ideally suitedlocation for wearable technology as it hampersmanualwork with the hands. An obvious alternative could beto measure at the wrist in a watch-type device, a

2

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

relatively unobtrusive location which is already in usefor the measurement of other health markers (Ruppand Balkin 2011, Valenti and Westerterp 2013, vanAndel et al 2015).

Yet, the measurement of PWV to the lower armsuffers from physiological factors that can influencethe BP relationship. In the simplified model presentedin equation (4) it is assumed that the arterial propertiesγ, r, E0 and h are constant. This assumption does nothold by default as the pulse propagates through a seriesof vessels with changing properties. This non-uni-formity in the vessel path increases as the pulse arrivallocation ismore distal.

Another effect is that the vessel radius h and thewall thickness r can be altered within peripheral arter-ies by muscle tissue in the arterial wall called smoothmuscles (Mukkamala et al 2015). Smooth muscles areactivated by the autonomic nervous system to controlblood flow. This is done in response to stressors suchas physical exercise or for thermoregulatory purposes.

Another limitation of the Moens-Kortewegequation is that it assumes the vessel through whichthe pulse propagates to be fixed in altitudewith respectto the heart. This only holds in very specific condi-tions.When the human is not limited inmobility, pos-ture changes can have significant impact on therelationship between PWV and BP. Attempts havebeen made to extend the Moens-Korteweg model toaccount for this (Poon et al 2006, McCombie 2008,Thomas et al 2015) but ideally posture change shouldbe avoided to ensure a reliablemeasurement.

Thus, ideally, the vessel over which PWV isderived should be (1) as fixed in altitude as possiblerelative to the heart, (2) contain aminimumof smoothmuscle tissue and (3) be as uniform as possible.Requirements two and three can be controlled byselecting a location that is relatively proximal to theheart, ensuring a short traversal path and little smoothmuscles. The first requirement can bemet by choosinga pulse arrival site on a limb that does not change inaltitude relative to the heart. This invites for the mea-surement of PWV to the head (He et al 2012) or to thechest (Solà et al 2009).

While measurement of PWV to a proximal loca-tion seems to alleviate many of the practical issues, italso has its drawback. In practice, SBP is always mea-sured at the brachial artery with a cuff wrapped aroundthe upper arm. As this is a relatively peripheral loca-tion the measured BP is also indicative of peripheralBP. When measuring PWV centrally, the obtained BPpredictions could therefore also be indicative of cen-tral BP and thusmay reflect central rather than periph-eral hemodynamics. In clinical practice this is alreadyan accepted parameter, in which case PWV is derivedfrom the time difference between the femoral and car-otid pulse wave arrival (femoral-carotid PWV). This isin contrast to the PWV that is defined in this paper,which is sometimes differentiated from femoral-car-otid PWV as R-wave gated PWV (Naschitz et al 2004).

There are known differences between central and per-ipheral BP, however it has also been shown that thetwo parameters can be predicted from each other to acertain extent (Karamanoglu et al 1993).

A final issue is the accuracy of the PWV measure-ment itself.When PWV ismeasured over a short arter-ial segment, small measurement errors in PAT canpropagate into large estimation errors of SBP. Thus, itcould be that the use of very long arterial segmentswould provide a more accurate measurement of PWVand thus reduce the sources of error. This could beachieved by measuring PWV over very long arterialsegments, with pulse arrival locations at the extre-mities of the body such as on the ankle. The ankle as apulse arrival location has been studied in the past forhemodynamic measurement (Nitzan et al 2002, Fooet al 2007, Ankle Brachial Index Collaboration 2008,Padilla et al 2009). It could be used to derive the Ankle-Brachial Index (revealing peripheral artery disease) orthe measurement of PWV in children (where othersensor locations might become inaccessible due tosmall limbs).

In summary, there are practical issues related toeach arterial site on which pulse arrival can be mea-sured. More proximal locations such as the headsatisfy the conditions of the Moens-Kortewegequations better but could potentially be insensitive toperipheral blood pressure variations, while very distallocations could reduce estimation error in pulse arri-val but may be confounded by a variety of factors. Theuse of the lower arm could be a middle-way betweenboth locations, but there has been no explorationwhe-ther the non-ergonomic finger clip can be replaced byawrist-worn device.

2. Research goals

The goal of this study is to show how the choice ofarterial site affects the obtained PWV and its relationto SBP. Specifically, the goal is to find out how PWV toproximal and very distal arterial sites perform inpredicting SBP, relative to PWVmeasured to thefinger(as the default location). The secondary goal is toconfirm that PWVmeasured to the finger is equivalentto PWV measured to the wrist. These tests help tounderstand whether it is feasible to measure PWV as asurrogate marker of BP in a more practical way thanusing a PPG clip on thefinger.

3.Materials

Twenty volunteers participated in the current trial.Three participants were excluded from the dataanalysis due to untrustworthy signals from the refer-ence device (CNAP blood pressure monitor 500,CNSystems), showing a high jump in blood pressurebefore and after recalibration. The remaining 17participants (mean age 31.4 years; 8 female) were

3

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

recruited within Philips Research. More detailedsubject characteristics are shown in table 1. Prior tothe start of the trial, participants received an oral andwritten explanation of the study procedure. Allparticipants provided written consent. The observa-tional protocol was approved by the internal ethicscommittee for Biomedical Experiments of PhilipsResearch Eindhoven in conformity with the declara-tion of Helsinki. Exclusion criteria included: sufferingfrom any chronic disease (such as diabetes, cardiovas-cular and pulmonary diseases), functional and cogni-tive impairments, use of medication affecting thehormonal, metabolic or cardiovascular system, preg-nancy and incapability to perform sport relatedexercises.

Subjects were seated in a normal chair behind acycle ergometer, positioned so that the full extensionof the knee was not reached during a complete cyclemotion. Their left arm was resting on a table, ensuringstability and armmuscle relaxation. Subjects were out-fitted with four PPG sensors; one on the right earlobe(TSD200, Biopac), one on the right index finger(TSD200, Biopac), one on the dorsal side of the rightwrist, proximal of the ulnar styloid process (PhilipsWeST, similar to the device validated in Valenti andWesterterp 2013) and an identical device on the distalside of the right ankle, proximal of the fibula styloidprocess. The Biopac sensors are attached with aspring-based clip, ensuring approximately the samecontact force regardless of the size of the finger andear. The Philips WeST devices were held in positionwith sweatbands. The experimenter ensured that thesweatbands were tight enough to keep the PPG sensorsin place without being too tight for the participant.Blood pressure was recorded using a CNAP bloodpressure monitor 500 (CNSystems), placing the cuffaround the left upper arm and the finger cuffs on the

index and middle finger of the left hand. All sensorswere connected to a Biopac acquisition system (Bio-pac) sampling at 1000 Hz ensuring synchronizedrecordings of the signals. A schematic overview of themeasurement setup is given infigure 1.

Participants were instructed to ensure a comfor-table position, keeping the right foot during the restperiods at the lowest point and returning to this pointafter each cycling session. The protocol was similar tothe one used in (Sun et al 2016a), starting with fiveminutes of rest followed by three sessions, each con-sisting offiveminutes cycling and fiveminutes rest.

4.Methods

4.1. Signal processingThe raw PPG signals were filtered. Baseline PPGmodulation due to respiration was removed with ahigh-pass Butterworth filter (cut-off= 0.4 Hz). Subse-quently, a low-pass filter was applied to remove high-frequency noise (cut-off= 10Hz).

After data cleaning, the quality of each heart beaton each PPG sensor was examined with a template-matching algorithm (Li and Clifford 2012). As theexperiment involves cycling, the ankle typically had ahigher rejection rate due to motion artifacts. The ear,being always at the same hight above the heart duringthe experiment, typically had the lowest rejection rate.As a compromise between quantity and quality, arejection threshold of 0.7 was used, which correspondsroughly to the rejection of the most intensive cyclingperiods. However, the gradual decrease of blood pres-sure just after exercise bouts is not rejected, whichensures a sufficient level of variation in BP for correla-tion analysis.

ECG R-peaks were localized with an enhancedvariation of the Hamilton-Thompkins QRS detector(Fonseca et al 2014).

For the detection of the pulse arrival, three possi-ble methods could be used (Rapalis et al 2014). Theseare (1) the foot of the PPG pulse, (2) its maximal risingslope or (3) the systolic peak. These are illustrated infigure 2. All three methods were used and one wasselected empirically based on correlation (seesection 4.2 and 5) for results. It is important to use thesame pulse arrival detector on all PPG signals as theselectedmethodwill affectmeasured PAT.

Using the computed PAT measurements, PWVwas determined, for which the PAT was adjusted forthe arterial segment length. The segment length persensor location can be estimated from the subjectheight using the known standard proportions of thehuman body, taken from (Winter 2009). This has beendone before for the estimation of PWV to the wrist in(Gesche et al 2012). A body correlation factor (BDC) isderived which expresses the length of an arterial path-way relative to total body height:

( )= * hPWV BDC PAT. Here, h is the person’s

Table 1.Participant characteristics.

Subject Age (years) Sex Weight (kg) Height (cm)

S1 42 Male 85 178

S3 27 Male 83 183

S4 24 Female 62 164

S5 24 Female 68 172

S6 53 Male 83 180

S7 25 Male 82.5 182

S8 32 Female 66 168

S9 28 Male 78 185

S10 22 Female 53 168

S11 22 Female 53 168

S12 58 Male 81.5 182

S13 27 Male 75 182

S16 27 Female 65 175

S17 31 Female 59 167

S18 35 Male 101 199

S19 28 Male 75 175

S20 28 Male 74 180

Average 31 41%F 73 177

4

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

height in meters (m) and PAT is in seconds (s), result-ing in PWV in m s–1. The BDC was 0.50 for the wrist,0.52 for the finger, 0.18 for the ear and 0.80 for theankle. These population averages do not take intoaccount the ape factor, which is an individual variationin arm and leg length from the general average. How-ever this is not a limitation for the trend analyses inthis study.

Finally, the continuous blood pressure measure-ment from the CNAP is used to determine SBP valuesfor each considered beat. For this, a single beat was iso-lated between two consecutive R peaks on the ECGand themaximumvaluewas taken as the SBP.

4.2. Statistical analysisFirst off, the different PPG markers for pulse arrival(see figure 2) are compared in terms of correlationwith SBP. The selected signal processing method isthen used to further study both the distribution ofPWV and its relation to blood pressure from onesensor location to another.

Once the signal processing method is selected, thecalibration procedure will be studied from subject tosubject. It was explained in the introduction that amapping between SBP and PWV is possible of theform ( )= * +a bSBP ln PWV . The parameter a isproportional to the artery-specific property γ, namely=

ga 2 (see equation (4)). It intuitively corresponds to

the responsiveness of BP to SBP as its unit is(mmHg ln m

s). Thus, the parameter a will be exam-

ined between different sensor locations. The focus ison understanding which of the sensor locations have amore stable a across subjects. Such a location couldpotentially suffer less from calibration problems.

Finally, using the calibrated regression functions,the PWV will be used to predict SBP for all subjects.The results are visualized per sensor location and sta-tistics are given about the accuracy of each location, interms of the correlation as well as the mean absoluteerror between predicted SBP and true SBP.

5. Results and discussion

5.1. PPGmarkers of pulse arrivalThe different PPGmarkers of pulse arrival were testedin terms of the number of beats for which the PPGmarker was successfully detected. Subsequently thecorrelation was computed per subject between SBPand the the PWV derived with that particular PPGmarker. In table 2 an overview is given of this analysisin terms of averages and standard deviation over thesubjects. In figure 3 boxplots are presented of theobtained correlations.

For all sensor locations the onset of the pulse wasdetected in more beats than any other marker of pulsearrival. The slope method resulted in the highest cor-relation with SBP for wrist, finger and ankle, while onthe ear the onset method showed a slightly better cor-relation with SBP. The peak-based PPG marker per-formed worst on all arterial sites: the peak was themost difficult to detect leading to the smallest numberof found beats and its correlation with SBP was lowest.

Figure 1. Schematic overview of themeasurement set-up and acquired signals. The signal of lead-II ECG (ground lead not shown)wasomitted.

Figure 2. Landmarks and time intervals visualized for acardiac cycle. ECG is in blue andwrist PPG is in red. The redcrosses denote (from left to right) the ECGRpeak, the PPGonset,maximum slope and peak. The horizontal black linesdenote the three pulse arrival time intervals.

5

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

Based on this, it was chosen to use the maximal slopeof the PPG pulse as an indicator pulse arrival as it per-forms verywell for all sensor locations.

5.2. Site-related differences in PWVand its relationto SBPThe PWV-SBP ratio a was derived per sensor locationfor all subjects. In figure 4 these calculated values of aare given. Infigure 5 the boxplots are shown of values afor all subjects for a certain sensor location.

The differences in a from one sensor location toanother were significant (using a paired samples t-test)for ear and wrist (p<0.0001), ear and ankle(p<0.0001), ear and finger (p<0.0001), finger andankle (p<0.0051) and ankle and wrist (p<0.01).The only exception was the comparison between thewrist and the finger, where no statistical differenceswere found.

It can also be observed in figure 4 that the para-meters a are correlated across subjects: people with a

small a in one sensor location also have the tendencyto have a small a on another sensor location. Pearson’scorrelation analysis shows that the wrist is highly cor-related with finger (c = 0.88) and ear (c = 0.85) andmoderately correlated with the ankle (c = 0.55). Theear and ankle are also correlated (c = 0.67) while theear and finger have a high correlation aswell (c= 0.89).

However, there are also clear differences betweensensor locations. The factor a is the highest for the ear:small changes in ear PWV come with large changes inSBP. Also the inter-subject variance is highest for theear (see also ear confidence interval in figure 5). On theother end of the spectrum, for subjects with an overalllower PWV, the values of a are very close to each otherregardless of the sensor location. For subjects S17, S3,S10 and S16 the differences in a from one sensor loca-tion are very small compared to the differencesobserved within subjects with high PWV such as S19,S9 and S13. These findings hint at the mediating effectof peripheral resistance on BP: at the proximal ear(with relatively little vasomotion in the pathway) there

Table 2.Averages and standard deviations, oversubjects, per PATmethod, for the number of beatsanalyzed (column 1) and the correlation of systolicblood pressure (column2)with the PATmethod.

# of beats Corr SBP

wrist onset 1670±421 −0.61±0.21wrist slope 1659±429 −0.66±0.18wrist peak 1649±429 −0.54±0.19ankle onset 1219±486 −0.42±0.27ankle slope 1214±486 −0.47±0.27ankle peak 1172±478 −0.38±0.32finger onset 1527±429 −0.63±0.20finger slope 1525±429 −0.65±0.19finger peak 1519±430 −0.53±0.23ear onset 1620±295 −0.65±0.16ear slope 1561±351 −0.60±0.21ear peak 1552±349 −0.46±0.20

Figure 3.Distribution of correlations over subjects for each ofthe PATmethodswith SBP.

Figure 4.The factor a over subjects per sensor location, wherea is a parameter fitted per subject in the function

( )= * +a bSBP ln PWV .

Figure 5.Boxplot per location of the calibration slope a.

6

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

is a large inter-subject variance but as the sensor loca-tion is placed more distally, it is observed that PWV isnormalized to a target level by the periphery, thusreducing inter-subject variance.

5.3. Prediction of SBPwith PWV to different arterialsitesIn table 2 the correlations between SBP and PWV fromall arterial sites is presented, however that correlationwas averaged over all subjects, concealing manydetails. Therefore, the calibration function for eachsubject was applied to obtain SBP from PWV. Theresulting SBP and PWV values are visualized infigure 6. Spearman’s correlation was 0.93 for all sensorlocations. In figure 7 the Bland-Altmann analyses aregiven that correspond to the same data. Mean errorsfor all sensors were 0 mmHg except for ear (meanerror of 1 mmHg). The confidence interval (CI) of theerror was respectively 11 mmHg, 10 mmHg,14 mmHg and 11 mmHg for wrist, finger, ear andankle.

The correlation between SBP and PWV is evident,though it is also observed that the predictions are notan exact match. In all sensor locations, a seeminglywhite noise is visible in the correlation analyses offigure 3. The Bland–Altmann figures show that this

noise has m » 0 mmHg and a 90% CI of 20 to26 mmHg. This noise could be attributed to the effectof breathing on the relationship between PWV andBP. The continuous rhythmic variations in intra-thor-acic pressure due to breathing cause modulations inheart rate and PWV. The variations in PWV have beenobserved in (Drinnan et al 2001) to be maximally 14ms to the finger. For comparison, in our study thePWV to finger was 2.6 m s–1 on average, thus a changeof 14 ms would impact PWV with 0.06 m s–1. Thiswould correspond to a 21 mmHg for the average sub-ject, which is close to the 90% CI of the finger of20 mmHg.

5.3.1.WristThe Bland-Altmann analyses for the wrist and fingerreveal comparable statistics: both have a small estima-tion bias of equal magnitude and a high correlation. Insection 5.2 it was also shown that the calibrationparameters are very similar between wrist and finger.Thus, the evidence suggests that the wrist PPG couldreplace the finger PPG with little risk. However, it isnot known which of the sensor locations is impactedmore by thermoregulation induced vasocontriction.

Figure 6.The relation between pulse arrival time to themaximal slope of the PPG and systolic blood pressure (per beat, for eachsensor). Each shade denotes a different subject. Pearson’s correlation (C) andmean absolute error (MAE) are reported in the titles ofthe figures.Half of the data points was chosen at random for visualization to increase readability.

7

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

5.3.2. EarWhile correlation analysis for the ear PWV yielded asimilar result as to wrist and finger, the Bland-Altmann figure reveals a higher error range caused bymore outlier points. This could be due to the earlierexplained differences between proximal and distallocations for pulse arrivalmeasurement.

5.3.3. AnkleThe PWV to the ankle becomes more noisy withhigher PWV levels. This is likely due to the motionartefacts during cycling, which increases PWV andSBP. This is clear in both figures 6 and 7. However, therelation is also slightly nonlinear: the response is flatterfor low SBP than it is for higher SBP. This is clearlyvisible in the Bland-Altmann analysis, where a linearerror trend is visible for each subject (each subjectdenoted by a distinct color). This stronger increasecould be because the legs are actively involved in thephysical activity. This causes a stronger increase inPWV to the leg than to other limbs due to localvasoconstriction. This might indicate that when usingPWV for the tracking of SBP during physical activity,the relationship between PWV and SBP could bealtered in the limb that is involved in the physicalactivity.

6. Conclusions

While PWV to all sensor locations does show signifi-cant correlation with SBP, there are clear advantagesand drawbacks associated with each location. The earPWV as a proximal location has its theoreticaladvantage of being less affected by vasomotion andorthostatic pressure changes, however a larger inter-subject variability was found in the calibrationbetween SBP and PWV. This should be accounted forin the design of an ear-based PWV measurementsystem. The wrist location, attractive for the wide-spread use of PPG in wrist watches, was shown to beequivalent in all tests to the finger PPG. This findingsuggests that the many positive results obtained forfinger PPG in the past are translatable to a wrist-basedmeasurement system, however also the drawbacks ofthe finger PPG sensor locationmight also be inherited,such as the strong effects of orthostatic pressure andvasocontriction. The effect of thermoregulatory vaso-constriction on the wrist remains to be studied. Theankle PPG, being the most distal sensor location, wasalready expected to perform less well as it can suffergreatly from the many hemodynamic variables overthe arterial path. The positive property of the anklehowever is that it is easier to accurately estimate PWVsince measurement errors in PATs propagate into

Figure 7.Bland-altmann analysis of the results presented infigure 6. Each shade denotes a different subject. Half of the data points waschosen at random for visualization to increase readability.

8

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

smaller PWV estimation errors than they do in moreproximal locations. This study suggests that the ankleis indeed too distal for measurement, however thesefindings should be interpreted with care as a fitnessintervention involving the legs was performed.

References

Ankle Brachial IndexCollaboration 2008Ankle-brachial indexcombinedwith FraminghamRisk Score to predictcardiovascular events andmortality: ameta-analysis J. Am.Med. Assoc. 300 197

Chase JG, StarfingerC, LamZ, Agogue F and ShawGM2004Quantifying agitation in sedated ICUpatients using heart rateand blood pressurePhysiol.Meas. 25 1037

Cutler J A, Sorlie PD,WolzM, ThomT, Fields L E andRoccella E J2008Trends in hypertension prevalence, awareness,treatment, and control rates in united states adults between1988–1994 and 1999–2004Hypertension 52 818–27

DrinnanM J, Allen J andMurrayA 2001Relation between heart rateand pulse transit time during paced respiration Physiol.Meas.22 425

ESH/ESC 2013 2013 ESH/ESC guidelines for themanagement ofarterial hypertension: TheTask Force for themanagement ofarterial hypertension of the European Society ofHypertension (ESH) and of the European Society ofCardiology (ESC)Eur. Heart J. 34 2159–219

Fonseca P, Aarts RM, Foussier J and LongX 2014Anovel low-complexity post-processing algorithm for preciseQRSlocalization SpringerPlus 3 1

Foo J YA, LimCS,Wilson S J,WilliamsGR,HarrisMA andCooperDM2007Pulse transit time ratio as a potentialmarker for paediatric crural and brachial blood pressureindex J. Hum.Hypertension 21 415–7

GescheH,GrosskurthD,Küchler G and Patzak A 2012Continuousblood pressuremeasurement by using the pulse transit time:comparison to a cuff-basedmethod Eur. J. Appl. Physiol. 112309–15

Greenland P et al 2010 2010ACCF/AHAguideline for assessment ofcardiovascular risk in asymptomatic adults J. Am. CollegeCardiol. 56 e50–103

HeDD,Winokur E S and Sodini CG 2012An ear-worn continuousballistocardiogram (BCG) sensor for cardiovascularmonitoring Proc. Annual Int. Conf. of the IEEE Engineering inMedicine and Biology Society, EMBS pp 5030–3

InanOT et al 2015 Ballistocardiography and seismocardiography: areview of recent advances IEEE J. Biomed. Health Inform. 191414–27

KaramanogluM,O’rourkeM, Avolio A andKelly R 1993Ananalysis of the relationship between central aortic andperipheral upper limb pressure waves inmanEur. Heart J. 14160–7

KortekaasMC,Niehof S P, vanVelzenMH,Galvin EM,Stolker R J andHuygen F J 2012Comparison of bilateral pulsearrival time before and after induced vasodilation by axillaryblockPhysiol.Meas. 33 1993

LiQ andCliffordGD2012Dynamic timewarping andmachinelearning for signal quality assessment of pulsatile signalsPhysiol.Meas. 33 1491

McCombieDB 2008Development of awearable blood pressuremonitor using adaptive calibration of peripheral pulse transittimemeasurementsPhDThesisMassachusetts Institute ofTechnology

Mukkamala R,Hahn J-O, InanOT,Mestha LK, KimC-S andHakanT 2015Toward ubiquitous blood pressuremonitoringvia pulse transit time : theory and practice IEEE Trans.Biomed. Eng. 62 1879–901

Naschitz J E et al 2004 Pulse transit time by R-wave-gated infraredphotoplethysmography: review of the literature and personalexperience J. Clin.Monit. Comput. 18 333–42

NitzanM,KhanokhB and Slovik Y 2002The difference in pulsetransit time to the toe and fingermeasured byphotoplethysmography Physiol.Meas. 23 85–93

Padilla JM, Berjano E J, Sáiz J, Rodriguez R and Fácila L 2009 Pulsewave velocity and digital volume pulse as indirect estimatorsof blood pressure: pilot study on healthy volunteersCardiovascular Eng. 9 104–12

PayneRA, Symeonides CN,WebbD J andMaxwell S R J 2006 Pulsetransit timemeasured from the ECG: an unreliablemarker ofbeat-to-beat blood pressure J. Appl. Physiol. 100 136–41

Peter L,NouryN andCernyM2014A review ofmethods for non-invasive and continuous blood pressuremonitoring: Pulsetransit timemethod is promising? Irbm 35 271–82

Pickering TG,Hall J E, Appel L J, Falkner B E,Graves J,HillMN,JonesDW,Kurtz T, Sheps SG andRoccella E J 2005Recommendations for blood pressuremeasurement inhumans and experimental animals: I. Blood pressuremeasurement in humans: a statement for professionals fromthe subcommittee of professional and public education of theAmericanHeart AssociationCouncil on high blood pressureresearchHypertension 45 142–61

PoonC, Zhang Y-T and Liu Y LY 2006Modeling of pulse transittime under the effects of hydrostatic pressure for cufflessblood pressuremeasurements 2006 3rd IEEE/EMBSInternational Summer School onMedical Devices andBiosensors pp 65–8

Proença J,Muehlsteff J, Aubert X andCarvalho P 2010 Is pulsetransit time a good indicator of blood pressure changesduring short physical exercise in a young population? 2010Annual Int. Conf. of the IEEE Engineering inMedicine andBiology Society (EMBC) (Piscataway,NJ: IEEE) pp 598–601

Rapalis A, Janusauskas A, Lukosevicius A andMarozas V 2014Noiseresistantmethod for cardiac pulsewave arrival timeestimationElektronika ir Elektrotechnika 20 59–62

RuppTL andBalkin T J 2011Comparison ofmotionlogger watchand actiwatch actigraphs to polysomnography for sleep/wakeestimation in healthy young adultsBehav. Res.Methods 431152–60

Smith RP, Argod J, Pépin J-L and Lévy PA1999 Pulse transit time:an appraisal of potential clinical applicationsThorax 54452–7

Solà J, Vetter R, Renevey P, ChételatO, Sartori C andRimoldi S F2009 Parametric estimation of pulse arrival time: a robustapproach to pulse wave velocity Physiol.Meas. 30 603

SPRINTResearchGroup 2015A randomized trial of intensiveversus standard blood-pressure controlN. Engl. J.Med. 3732103–16

Sun S, Bezemer R, LongX,Muehlsteff J andAarts RM2016aSystolic blood pressure estimation using PPG and ECGduring physical exercise Physiol.Meas. 37 2154

Sun S, Bezemer R, LongX,Muehlsteff J andAarts RM2016bSystolic blood pressure estimation using ppg during physicalexerciseProc. Annual Int. Conf. of the IEEE Engineering inMedicine and Biology Society, EMBS p 2415

TangZ, Tamura T, SekineM,HuangA, ChenW, YoshidaM,Sakatani K, KobayashiH andKanaya S 2016A chair-basedunobtrusive cuffless blood pressuremonitoring systembasedon pulse arrival time IEEE J. Biomed.Health Inform. 99 1

TengX-F andZhang Y-T 2007Theoretical study on the effect ofsensor contact force on pulse transit time IEEETrans.Biomed. Eng. 54 1490–8

Thomas S S,NathanV, ZongC, SoundarapandianK, Shi X andJafari R 2015 BioWatch: a non-invasive wrist-basedbloodpressuremonitor that incorporates training techniquesforposture and subject variability IEEE J. Biomed.HealthInform. 20 1291–3000

U.S. Food andDrugAdministration 2015GeneralWelness Policy forLowRiskDevices—Draft Guidance for Industry and Food andDrugAdministration Staff

Valenti G andWesterterp K 2013Optical heart ratemonitoringmodule validation study 2013 IEEE Int. Conf. on ConsumerElectronics (ICCE) (Piscataway,NJ: IEEE) pp 195–6

9

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al

vanAndel J, UngureanuC, Aarts RM, Leijten F andArends J 2015Using photoplethysmography in heart ratemonitoring ofpatients with epilepsy Epilepsy Behav. 45 142–5

Wijshoff R,MischiM andAarts RM2016Reduction of periodicmotion artifacts in photoplethysmography IEETrans.Biomed. Eng.

WinterDA2009Biomechanics andMotor Control of HumanMovement (NewYork:Wiley)

WongMY-M, PoonCC-Y andZhang Y-T 2009An evaluation ofthe cuffless blood pressure estimation based on pulse transit

time technique: a half year study on normotensive subjectsCardiovascular Eng. 9 32–8

YoonG, Lee J Y, JeonK J, ParkK-K, YeoH S,HwangHT,KimHS andHwang I-D 2002Multiple diagnosis based onphotoplethysmography: hematocrit, spo2, pulse, andrespiration Photonics Asia 2002 (International Society forOptics and Photonics) pp 185–8

ZhangG,GaoM,XuD,Olivier NB andMukkamala R 2011 Pulsearrival time is not an adequate surrogate for pulse transit timeas amarker of blood pressure J. Appl. Physiol. 111 1681–6

10

Biomed. Phys. Eng. Express 3 (2017) 015022 MRadha et al