IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO.

9, SEPTEMBER 2010 2391

Instruments and Methods for Calibration of

Oscillometric Blood Pressure Measurement Devices
Eulalia Balestrieri and Sergio Rapuano, Senior Member, IEEE

Abstract—Most of automatic blood pressure measurement de- are already employing automated technology to measure BP
vices on the market implement an oscillometric method. This instead of the traditional mercury sphygmomanometers [1].
paper highlights the need for standardized calibration procedures Most of the automated noninvasive BP (NIBP) measurement
for these devices in order to reduce logistical difficulties, time-
consuming processes, and high costs involved in clinical vali- devices for clinical use rely on oscillometric techniques. Even
dations. A critical overview of researchers’ and manufacturers’ if the basic principle is the same as the Korotkoff method,
proposals to answer this need is presented as a contribution to the the cuff pressure is measured electronically, during inflation
IEEE P1721 project for a new standard on Objective Measure- or deflation phases, and used to estimate SBP and DBP by
ment of Systemic Arterial Blood Pressure in Humans. means of proprietary algorithms. Unfortunately, systematic
Index Terms—Blood pressure (BP), calibration, noninvasive uncertainty components affect oscillometric BP measurement
measurement, oscillometric method, simulators, standardization. devices (OBPMDs), too. In particular, they can be affected
both by lack of calibration of the electronic pressure sens-
I. I NTRODUCTION ing system and by the uncertainty due to the DBP and SBP
estimation algorithm. As the algorithms are confidential and
T HE IMPORTANCE of accurate blood pressure (BP) mea-
surement in the diagnosis of arterial hypertension and
in the antihypertensive treatment is largely acknowledged in
differ among instruments, clinical trials are adopted to validate
OBPMDs against manual auscultatory measurements carried
clinical practice [1]. out by applying the Korotkoff method [2]. Due to the scarce
The classic approach to BP measurement is based on use reproducibility of such calibration, adopted in all the inter-
of the mercury sphygmomanometer and the Korotkoff method. nationally available standards, OBPMDs can pass validation
A compression cuff wrapped around the patient’s arm is first tests despite producing clinically significant differences that
inflated until the circulation stops. Then, the cuff is slowly can be greater than 15 mmHg in some individuals [2]. This
deflated while the doctor listens at the sounds made from blood problem has to be considered even more serious since system-
flowing again in the patient’s artery by means of a stethoscope. atic uncertainty components in BP measurements caused by
Although it has been regarded as the gold standard for BP inadequate OBPMD calibration are a common cause of over-
measurement in a clinical setting for a long time [1], this and underdiagnose of hypertension [2].
method is known to be affected by a number of limitations Aware of this situation, the IEEE I&M Society TC-25 Sub-
due to measurement conditions, arm position, and observer’s committee on “Objective Blood Pressure Measurement” has re-
attitude. cently started the project P1721 “IEEE Standard for Objective
The observer-related biases could, in theory, be eliminated Measurement of Systemic Arterial Blood Pressure in Humans”
by using automated electronic devices, due to their capability of [3]. The standard in process aims to provide a comprehensive
providing printouts of systolic, diastolic, and mean BPs (SBP, guide to assess the uncertainty of automatic BP devices in an
DBP, and MBP, respectively) together with heart rate, time, objective way, independently from a group of human subjects
and date of measurement. Such an approach could eliminate or a specific clinical protocol [4].
interpretation errors, observer bias, and terminal digit prefer- A brief overview of the existing standards and protocols
ence [1]. Moreover, it allows 24-h monitoring of the patients’ aimed at the automated NIBP measurement device calibration
health with fixed time lapses. These potentialities, together has been introduced in [4] as a contribution to the new IEEE
with the capability of storing data for later analysis and/or standard. In particular, the carried-out analysis has highlighted
for remote transmission, promoted the diffusion of electronic that the device evaluation and calibration is mainly carried
BP measurement devices. A number of large research studies out by taking as a reference the measurements obtained on a
large number of people by two observers using stethoscope
and mercury sphygmomanometer. Moreover, the methods to
Manuscript received July 20, 2009; revised March 11, 2010; accepted do that do not always agree each other, producing questionable
April 17, 2010. Date of publication July 12, 2010; date of current version results [4].
August 11, 2010. The Associate Editor coordinating the review process for this A general description of the oscillometric BP methods has
paper was Dr. Salvatore Baglio.
The authors are with the Department of Engineering, University of been provided in [5]. In particular, the differences that can
Sannio, 82100 Benevento, Italy (e-mail: balestrieri@unisannio.it; rapuano@ be found among different devices implementing, in different
unisannio.it). ways, the same basic idea in terms of general specifications and
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. accuracy have been carried out. The most promising approach
Digital Object Identifier 10.1109/TIM.2010.2050978 to a reproducible calibration of NIBP devices, i.e., to use

0018-9456/$26.00 © 2010 IEEE

2392 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

according to the oscillometric method is graphically shown

in Fig. 1 [6]. The basic principle of BP measurement using
the oscillometric technique depends on the transmission of
intra-arterial pulsation to the occluding cuff surrounding the
limb. A measurement relying on this basic principle could
be done by placing a cuff around the upper arm and rapidly
inflating it to about 30 mmHg above the SBP, thus occluding
the blood flow in the brachial artery [Fig. 1(a)]. The pressure
could then be gradually decreased, often in steps, such as
5–8 mmHg [Fig. 1(a)]. The pressure in the cuff should be
measured by a sensor and digitally recorded by means of a
data acquisition system. The resulting oscillogram should be
detected and processed at each pressure step. The cuff pressure
can also be deflated linearly in a similar fashion as per the
conventional auscultatory method. Arterial pressure oscillations
are superimposed to the cuff pressure when the blood vessel is
not fully occluded. Separation of the superimposed oscillations
from the cuff pressure is accomplished by filters that extract the
corresponding signals [Fig. 1(b)]. Signal sampling is carried out
at a frequency determined by the pulse or heart rate [7].
In the case of conventional oscillometric method, the pulsa-
tion measured with the cuff pressure sensor during the deflation
phase ideally starts at SBP and ends at DBP, as shown in
Fig. 1(a) [8], because before the SBP pulse, no blood flows in
the vessel, and after DBP, almost only the cuff pressure can
be measured. However, even when cuff pressure is greater than
SBP, it is impossible to occlude completely the brachial artery
Fig. 1. BP determination according to the oscillometric method [6]. (a) Cuff and stop arterial pulsation of the cuff as the lateral part of the
positioning and deflation. (b) Oscillogram obtained by removing the cuff cuff has lower pressure than the center. Therefore, it is not easy
pressure.
to determine the SBP point [8].
sources that emulate the BP waveforms coming from one or For these reasons, the oscillation amplitudes are most often
more physiological databases instead of groups of patients, has processed with an empirical algorithm to estimate SBP and
been finally introduced, discussing the main issues of such DBP [7]. In particular, mathematical criteria are applied to the
devices [5]. envelope of the waveform obtained by plotting the oscillatory
In this paper, the analysis in [5] is expanded, focusing on pulses versus the cuff base pressure. The baseline-to-peak
the main characteristics that an OBPMD calibration method amplitude, the peak-to-peak amplitude, or a quantity based
and instrumentation should have. In the next section, a gen- on the partial or full time integral of the envelope waveform
eral description of the oscillometric BP methods is provided. can be used as oscillometric indexes [7]. In Fig. 2, the main
The following three sections deal with the specifications and criteria to obtain the BP values from the envelope are shown.
accuracy differences that can be found among different devices Usually, the cuff pressure value corresponding to the envelope
implementing in different ways the same basic idea. Section VI maximum height (HM in Fig. 2) is considered the MBP. The
describes the most promising approach to a reproducible cal- main methods of determining the SBP and DBP are the height-
ibration of OBPMDs, which adopts sources emulating the BP and slope-based ones. In the height-based approach, which is
waveforms that come from one or more physiological databases the most used, the systolic and diastolic values are obtained as
instead of groups of patients. The main issues of such devices the cuff pressures corresponding to fixed ratios of the envelope
are two: how to provide a suitable pressure waveform for a amplitude to its maximum value (HS /HM and HD /HM in
given BP value, and which waveforms should be generated Fig. 2) [9]. Some authors proposed to take always the fractions
to reproduce the human BP signal in the most accurate way. of 40% and 60% out of the maximum amplitude to find SBP
An overview of the researchers’ and manufacturers’ proposals and DBP, respectively; thus, the oscillometric method could
in the field and the most interesting new trends is reported in be standardized [10], [11]. However, other authors reject such
Section VI. criterion due to the wide number of factors influencing the
oscillation amplitudes, including the different types of available
inflatable cuffs [12]. On the other hand, a study carried out
II. O SCILLOMETRIC M ETHOD
by Geddes [13] observed that the best correlated values with
Automated sphygmomanometers applying the oscillometric the auscultatory method are 50% and 80% for the systolic and
method were developed, for the first time, in 1876. They diastolic, respectively [9]. Drzewiecki [14] reported the values
were able to determine the SBP and DBP values by means of 0.593 and 0.717, respectively, using a theoretical model and
of mathematical algorithms. The SBP and MBP determination a constant input pressure amplitude.
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2393

teristics of differences in those implementations include the

following [22]:
1) different cutoff frequencies for the filtering of the pres-
sure waveforms;
2) different metrics for measuring pulse size;
3) different methods for averaging pulse size;
4) different methods for finding the peak of the oscillation
envelope;
5) different methods for determining the average cuff pres-
sure on a given pressure step;
6) different methods and coefficients for extracting systolic
and diastolic pressures from the oscillation envelope;
7) different methods of controlling cuff pressure such as step
down, step up, bleed down, and bleed up.
Since the oscillometric signal is complex and largely chang-
ing from patient to patient, the listed design alternatives can
lead to similar results on some patients and quite different
results on others. The design of an OBPMD is usually based
on the data measured on a particular group of patients [22].
Of course, the choice of subjects is up to manufacturers. In
theory, the groups are large enough that they should be statis-
tically identical and represent the whole population of patients.
In practice, they can still provide biased references. Trained
nurses are enrolled by the manufacturers to obtain the manual
pressure measurements to be used as references. The nurses are
supposed to adopt the same standard method, but in practice,
Fig. 2. Main criteria for oscillometric BP determination [6]. the process is still subjective, and different nurses get slightly
different results [4]. The final outcome is that, when they arrive
Comprising only two spectral components, the model in [14] on the market, the OBPMDs do yield similar average results,
did not consider the possible spectral diversity of the BP signal. while on any specific subject, the results can be quite different
Ursino and Cristalli [14] used a real BP pulse wave in their [22]. For this reason, through the years, both standards and
mathematical study and obtained ratios of 0.52 and 0.70 for clinical trial protocols have been published concerning the cal-
SBP and DBP values, respectively. However, the ranges of ibration of such devices. Each standard or protocol determines
the ratio deviations were quite wide, introducing considerable the device performance and accuracy in its own way [4]. The
errors [14]. required recruitment of a large number of volunteers, as well
The slope-based approach applies the derivative of the os- as the extensive range of BPs and of arm circumferences to be
cillation envelope with respect to cuff pressure [15]. The max- included in the trials, makes validations expensive to conduct.
imum and minimum of this derivative denote SBP and DBP, Therefore, it is unlikely that all devices on the market will
respectively [16], [17] (Fig. 2). be validated according to all the standards; therefore, their
Other algorithms have also been proposed for BP estimation characteristics cannot be compared quantitatively [23].
[18]–[20]. However, there is not a unique implementation of the
oscillometric method, as different OBPMD manufacturers use IV. OBPMD ACCURACY
different legacy implementations [21]. Most of the automatic
OBPMDs have been the subject of several research studies in
device manufacturers do not reveal the exact way in which their
order to achieve improvements in their measurement accuracy.
equipment calculates the pressure values.
One limitation of this kind of instruments is the empiricism
of the oscillometric algorithms, so the main contributions from
literature deal with the metrological compatibility of the mea-
III. NIBP D EVICE D IFFERENCES
sures provided from different instruments. Many other papers
The first companies in the oscillometric NIBP market deal with the influence factors on the measurements, such as the
patented their technology and tried to exert some control on measurement environment or the patient’s position and health.
the market by aggressively pursuing patent litigation. As a Among all OBPMD research contributions, two main di-
result, manufacturers successively entered in the market of rections can be identified: one focused on repeatability and
oscillometric NIBP devices were forced to use slightly different reliability facet; the other one on methods and indicators.
oscillometric method implementations in order to avoid law- Concerning the first research direction, individual devices
suit and to survive on the market. Therefore, currently, there have been shown to be sufficiently repeatable for clinical use,
are several implementations of the basic oscillometric method but differences among devices are sometimes so large that they
different enough not to violate any patent. The main charac- may be misinterpreted as clinically significant [24]. In [24],
2394 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

subjects with arrhythmia, atherosclerosis, and hypertension,

respectively, are shown [26].
Moreover, the measurement conditions, including patient
posture, arm position, cuff size, and environment in which BP
measurements are taken, are factors to be carefully controlled
in order to improve the OBPDM measurement accuracy [27],
since the consecutive BP measurements in the same individual
vary significantly, whether the measurements are taken manu-
ally or automatically [27]. Other disturbances, including heart
rate changes, frequent ectopic beats, respiratory disturbances,
coughing, talking, and muscle tension, associated with changes
in the oscillometric pulses in the cuff pressure for the oscillo-
metric technique, can also influence clinical BP measurement
variability [27].
The oscillometric measurement is also susceptible to noise
such as undesired external pressure applied to the cuff, result-
ing in the distorted oscillation [28]. The distorted oscillation
contributes to the BP measurement error. The scheme mini-
mizing the BP measurement error consists of noise detection
method and noise reduction method. The conventional noise
detection method uses a certain criterion to detect distorted
oscillation. However, the criterion is fixed at all times so that
the method may fail to detect the distorted oscillation in some
cases. The conventional noise reduction method clamps the
distorted oscillation to baseline or uses linear interpolation [28].
Therefore, there exists a possibility of losing the oscillation
information if the oscillation overlaps with the noise. The loss
of the oscillation information would cause the BP measurement
error [28].
Fig. 3. (a) Oscillation waveform from a subject with normal cardiovascular Among the research works dealing with the methods and
function. (b) Oscillation waveform from a subject with cardiac arrhythmia. indicators, the influence of different representations of the
(c) Oscillation waveform from a subject with atherosclerosis. (d) Oscillation
waveform from a subject with hypertension [26]. oscillometric index (defined as an envelope of a certain char-
acteristic physical property of data obtained with the OBPMD)
the results of a study carried out on 19 low-cost automated on the known height- and slope-based empirical algorithms for
OBPMDs using a repeatable artificial arm simulator are re- automatic determination of SBP and DBP has been evaluated
ported. In particular, the study looked for the repeatability of in [29]. In particular, in addition to the arterial pressure pulses,
the measures coming from the same device, called within- a typical physical property has been used for the oscillometric
device repeatability, and of the differences among measures index. Other properties such as a time derivative and an audible
coming from different devices, called between-device repeata- part of data measured by a microphone implanted in the cuff
bility. The devices were found repeatable (with average within- (Korotkoff sounds) have also been used in this study, carrying
device difference of 1 mmHg), but between-device differences out three different representations of oscillometric index curves
were 4.4 mmHg (SBP) and 3.6 mmHg (DBP), for normal and (Epp , Epc , and Epd ) for each type of physical property, using
high–normal BP values [24]. minimal (MIN), center of gravity (COG), and maximal (MAX)
The following are the factors that limit OBPMD accuracy: points for each heartbeat, as shown in Fig. 4. These represen-
1) The motion of the arm can induce noise that is interpreted tations have been evaluated on 92 measurements performed on
as pressure oscillation; 2) the arrhythmias can distort the os- 23 healthy subjects. Estimations of SBP and DBP obtained by
cillometric envelope; 3) the reliability and validity studies on the height- and slope-based algorithms were compared with the
OBPMDs have been conducted on selected populations only reference SBP and DBP measured with the Welch Allyn OSZ4
(by age and disease); 4) the studies tested the manufacturer’s Self-Measurement Blood Pressure System [29].
cuff only [19]; and 5) the epidemiologic data for cardiovascular It has been found that the height-based method can be applied
risk prediction are lacking [20], [25]. to envelopes that exhibit a fast amplitude change corresponding
Disturbances resulting from tremor or cardiovascular ab- to DBP and SBP. The same characteristic is also required for
normalities occurring during the measurement period greatly the slope-based method, but additional constraints should be ap-
distort the original oscillation amplitudes, leading to a change plied due to the nonmonotonic shape of envelopes (particularly
in the specific oscillation amplitudes that correspond to the SBP at SBP). For example, to determine SBP, the MAX slope of Epp
and DBP. In Fig. 3, the different real oscillation waveforms at a pressure level in the cuff has been constrained to assume
of normal and abnormal cardiovascular functions, correspond- a value of at least 15 mmHg above MBP and considering
ing to a normal subject without cardiovascular disease and values of the normalized oscillometric index less than 0.6. To
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2395

TABLE I
R ESULTS O BTAINED FOR S LOPE -BASED (W ITH AND W ITHOUT
C ONSTRAINTS A DDED ) AND H EIGHT-BASED M ETHODS FOR THE T HREE
D IFFERENT R EPRESENTATIONS OF O SCILLOMETRIC I NDEX C URVES OF
THE A RTERIAL P RESSURE P ULSES . C I S THE T OTAL N UMBER OF C ASES
C LASSIFIED AS G OOD ( UP TO 5-mmHg) D IFFERENCE B ETWEEN THE
E STIMATED AND THE M EASURED SBP AND DBP), F IS THE N UMBER OF
FAILED C ASES ( A D IFFERENCE OF A BOVE 10 mmHg), r IS THE L INEAR
R EGRESSION C ORRELATION, Δp ± SD IS THE M EAN D IFFERENCE AND
I TS S TANDARD D EVIATION IN mmHg, Epp IS THE MIN- TO -MAX
E NVELOPE OF THE O SCILLOMETRIC P ULSE, Epc IS THE C ENTERED OR
MIN- TO -COG E NVELOPE OF THE O SCILLOMETRIC P ULSE , AND Epd IS
Fig. 4. Three different oscillometric index (Epy ) representations of an os- THE COG- TO -MAX E NVELOPE OF THE O SCILLOMETRIC P ULSE [29]
cillometric pulse, where y is the envelope type, p is the pulse amplitude or
the MIN value of data to the MAX value of data for each pulsation, c is the
centered or MIN value of data to the COG value for each pulsation, and d is
the difference between p and c or the COG value to the MAX value of data for
each pulsation [29].

determine DBP, the search for pressure has been constrained

at least 2 mmHg below MBP and for normalized index less
than 0.9. In Table I, results obtained for slope-based (with and
without constraints added) and height-based methods for the
three different representations of oscillometric index curves of
the arterial pressure pulses are shown. For both methods, the
best results have been obtained for oscillometric index formed
from peak-to-peak values of pressure pulses. The results for
the height-based method show that, for Epp , there were no
failed cases and that most of the 92 cases were classified as
good, i.e., 87 for SBP and 86 for DBP. Concerning the slope-
based method, better results for SBP were obtained by using
the method with constraints. In that case, for Epp , 82 cases
have been classified as good and having only five failed cases.
Only 48 cases classified as good and 35 failed cases have been
obtained applying the SB method without constraints. On the
other hand, constraints do not improve the DBP results. For Epp
and Epc , slightly better results without constraints have been
carried out. For Epd , slightly better results have been carried
out with constraints.
Although SBP and DBP are most often measured in the
clinical environment, MBP, i.e., the average pressure during
a cardiac cycle, has particular importance in some situations,
because it is the driving force of peripheral perfusion. SBP
and DBP can vary significantly throughout the arterial system,
whereas MBP is almost uniform in normal situations [7]. of physiological variable of BP than sphygmomanometers do.
Although the measured MBP is reported by most OBPMDs, If oscillometrically measured BP does mark a different kind of
some researchers do not use it. Instead, they calculate the MBP physiological variable, this would be masked by adjusting the
from the SBP and DBP displayed by the device with a particular algorithms to mimic sphygmomanometer outcome, in order to
formula. Moreover, some devices do not report MBP [30]. In achieve an “A” grading in validation [30].
[30], measured and calculated MBPs obtained by two different However, many OBPMDs on the market provide BP read-
OBPMDs in two different patient groups have been analyzed, ings significantly different when compared with auscultation
finding a small but significant difference between the measured (Korotkoff sounds) [31]. Particularly with people who suffer
and calculated MBPs depending on age, SBP, and DBP, as well from cardiovascular diseases, the resulting accuracy is much
as on the OBPMD. The authors suggest researchers using an lower when compared to the auscultation method. Surprisingly,
oscillometric device to be aware of the difference in calculated there are no guidelines for testing oscillometric algorithms in
and measured MBPs. Moreover, they open some questions special cases such as patients with cardiovascular diseases [31].
about the comparison between oscillometric and sphygmo- Other than the motion artifact, OBPMDs are affected by
manometer BP measurement devices done by validation pro- the stepped deflation artifact [32]. When using oscillometry,
tocols as the Association for the Advancement of Medical there are two practical ways of deflating the cuff, namely, step
Instrumentation (AAMI) and the British Hypertension Society and linear deflations. The problem of linear deflation methods
(BHS) ones. This should be convenient when using guide- is that, at lower pulse rates, they can produce oscillometric
lines based on sphygmomanometer readings, but Kiers et al. profiles that are undersampled and can contribute to reduce the
[30] cannot exclude that oscillometry measures a different kind measurement accuracy, and at higher pulse rates, oversampling
2396 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

TABLE II
D ISTRIBUTION OF M EASUREMENT E RRORS FOR S EVERAL AUTOMATIC BP D EVICES [34]

extends determination time which can contribute to patient population mean error, which would be no more than 5 mmHg
discomfort. With stepped deflation, the deflation rate is based for either systolic or diastolic pressure. If this assumption is
on the patient’s pulse rate, which results in a constant number false, and the errors tend to cluster within persons (i.e., the
of samples for all pulse rates—no more or no less than the spread of the errors within a single person is less than that
number needed for an accurate and comfortable determina- for the entire population), then as more readings are gathered
tion. However, unfortunately, the stepped-deflation oscillomet- on a particular person, the average error will converge toward
ric measurements are contaminated with the stepped deflation that person’s true error. If that person’s error is more than
artifact arising mainly from bleed valve switching [32]. 5 mmHg, the monitor will do a poor job of providing an accu-
The OBPMD accuracy is usually classified as static and dy- rate average for that person, no matter how many measurements
namic. The static accuracy represents the accuracy with which are taken [34].
the internal pressure gauge can measure constant values of the In order to empirically estimate the proportion of people
cuff pressure. The dynamic accuracy represents the accuracy of for whom a validated NIBP measurement device produces
the device in taking actual readings on patients [22]. inaccurate readings, the raw data from three published val-
Testing of OBPMDs is currently performed by two methods. idation studies on both auscultatory and oscillometric auto-
The first method adopts a commercial test equipment man- mated monitors, according to AAMI and BHS protocols, have
aged by biomedical technicians. There are several commercial been examined in [34]. Both auscultatory and oscillometric
NIBP measurement device test instruments on the market. automated monitors were assessed. In particular, the examined
Such instruments usually feature static functions and simulate auscultatory monitors were the Schiller BR-102 produced by
oscillometric waveforms [23]. The second method is directed Schiller AG, operating in auscultatory mode, and the CH-Druck
to assess the dynamic OBPMD accuracy. In this second ap- produced by Disetronic Medical Systems. The examined oscil-
proach, the characterization is based on the comparison of the lometric monitors were instead the Schiller BR-102, operating
BP measurements carried out by means of the device under in oscillometric mode, and the Profilomat II produced by Dis-
test and those performed with a sphygmomanometer and a etronic Medical Systems. In Table II, the obtained distribution
stethoscope. Several guidelines have been developed to ensure of measurement errors (that are the differences between the
process reproducibility, usually requiring recruitment of a large monitor and observer readings) is shown. In particular, the
number of volunteers, extensive range of BPs, and of arm CH-Druck and Schiller BR-102 instruments for both SBP and
circumferences [4], [23]. Although a protocol, coming from the DBP, as well as the Schiller BR-102 using the oscillometric
European Society of Hypertension (ESH), has been developed, mode for DBP only, were within the validation criteria. For
requiring a limited number of volunteers (33), a comparison those NIBP measurement devices that passed the standard test
of the developed protocols on statistical basis has shown that thresholds, 19%–36% of the individual readings had errors
the reduced sample size results in a reduction of statistical outside the range of −5 to +5 mmHg, while 4%–10% had
power which brings into question the applicability of the ESH errors exceeding 10 mmHg in magnitude [34]. For the monitors
protocol [33]. that passed the validation criteria, 20%–38% of individuals
Moreover, also NIBP measurement devices that meet the had average measurements that were inaccurate (by more than
validation criteria specified in the international standards and 5 mmHg), and 2%–5% of individuals had average measure-
protocols may still be inaccurate as discussed in [34]. In partic- ments that were inaccurate by more than 10 mmHg. Analysis
ular, the analyses carried out by these validation criteria assess of the individual errors, using repeated measure analysis of
only the distribution of errors across the specified number variance, yielded estimates of between-person variance that
of individual readings, not the distribution of average errors accounted for 26%–63% of the total variance. This indi-
observed on each of the recruited participants. Implicitly, this cated that the errors of the measurements do tend to cluster
approach assumes that there is no between-person variance, i.e., within persons. From the obtained results, the authors affirm
no tendency for the monitor to be more accurate for some per- that the current validation standards for ensuring accuracy
sons and less accurate for others. This means that the average are inadequate to ensure that individuals receive accurate BP
error for a single person is assumed to converge toward the measurements [34]. Moreover, a two-stage validation is
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2397

proposed, first, by validation at the population level and, sec- Moreover, the accuracy of measurements must be traced to
ond, by validation in the physician’s office for the intended user. an accepted measurement reference source or standard. This is
the requirement of “traceability” that, according to the VIM, is
V. NIBP T EST S IMULATORS the property of the result of a measurement or the value of a
standard whereby it can be related to stated references, usually
As discussed in [4], a unified approach for the calibration
national or international standards, through an unbroken chain
procedure with acceptable execution time and costs is still
of comparisons, all having stated uncertainties.
missing and required. Since BP simulators that regenerate os-
Therefore, metrological calibration main purposes are the
cillometric waveforms promise an alternative to clinical trials,
following: 1) to ensure the readiness of test equipment to
provided that they include sufficient physiological and patho-
perform accurate measurements and provide valid data; 2) to
logical oscillometric waveforms, they could possibly upgrade
limit the number of erroneous test decisions; and 3) to maintain
or replace the current test methods of the OBPMDs [23].
overall measurement integrity and traceability. This involves
Research work [23], [35]–[37] seems to support the future
the following requirements:
replacement of clinical study validation in favor of simulator
evaluation. In order to reach the status of standard, however, 1) a clearly defined particular quantity that has been
the NIBP simulators should overcome several limits in terms of measured;
differences in reproducing the BP values. A possible solution 2) a complete description of the measurement system or
could be a general agreement about the waveforms to generate working standard used to perform the measurement;
and to take them from physiological databases. Therefore, 3) a stated measurement result or value, with a documented
the “ideal” simulator should be provided by a comprehensive uncertainty;
database of physiological waveforms, a validation protocol that 4) a complete specification of the stated reference at the time
assesses its accuracy over comprehensive range of conditions that the measurement system or working standard was
and its independent verification [35]. compared to it;
In the following, the measurement requirements for NIBP 5) establishing the status of the stated referencing at the time
test simulators to be a reliable reference for calibrating that the measurement system or working standard was
OBPMDs, the current situation of NIBP test simulators as compared to it [39].
well as some of the last research contributions concerning
physiological waveform databases, modeling, and simulation The main function of an oscillometric NIBP simulator is to
are presented. dynamically reproduce the pressure profile of a live subject
during a BP measurement cycle [40]. This seems making
A. NIBP Test Simulator as a Reference Standard them capable of possibly upgrading or replacing the current
test methods of the NIBP measurement devices, removing
Metrological calibration involves the process of measure- their serious drawbacks. However, most commercial simulators
ments and comparisons of a measurement standard to either generate artificial waveforms that significantly differ from the
another measurement standard or a device of unknown accuracy physiological ones, making these devices usable only for NIBP
(unit under test). According to International Vocabulary of monitor static calibration. Moreover, different NIBP simulator
Metrology (VIM) [38], calibration is the set of operations that manufacturers use different ways to create the pulse pressure
establish, under specified conditions, the relationship between envelopes.
the values of quantities indicated by a measuring instrument
and the corresponding values realized by standards. Impor-
tant information to be assessed in this operation concerns the B. Current NIBP Test Simulators
following. Commercial NIBP test simulators were introduced, for the
1) Accuracy—“closeness of agreement between a mea- first time, in the early 1990s for testing and evaluating oscil-
sured quantity value and a true quantity value of a lometric BP monitors. Two different types of simulators have
measurand” [38]. been developed since then: the artificial limb and the waveform
2) Precision—“closeness of agreement between indications generator. Limb simulators consist of an artificial limb, usually
or measured quantity values obtained by replicate mea- an arm, which incorporates an artificial artery with pulsating
surements on the same or similar objects under specified fluid. In order to test the OBPMD, its cuff is wrapped around
conditions” [38]. the artificial arm. Waveform simulators generate oscillometric
3) Repeatability—“condition of measurement, out of a set waveforms fed to the pneumatic hose of the OBPMD under test.
of conditions that includes the same measurement pro- The latter type of simulators is predominantly used today [41].
cedure, same operators, same measuring system, same However, unlike limb-type simulators, waveform simulators are
operating conditions, and same location, and replicate not capable of taking into account the transfer function between
measurements on the same or similar objects over a short arterial pressure in the limb, air pressure in the cuff, the effects
period of time” [38]. of cuff bladder size, placement of the cuff, and the material
4) Reproducibility—“condition of measurement, out of a set from which the cuff is made of. These variables could only be
of conditions that includes different locations, operators, assessed by performing clinical validation on human subjects,
and measuring systems, and replicate measurements on since currently available commercial waveform simulators are
the same or similar objects” [38]. not able to replay realistic human signals [42].
2398 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

Fig. 5. NIBP monitor evaluation protocol [46].

Several NIBP test simulators have been developed, and their

operating modes are described in [43]. There are also many
reported studies dealing with the NIBP device calibration by
using a simulator, mainly devoted to highlight the advantages
that a similar calibration can provide [36], [37], [44].
An evaluation protocol using NIBP test simulators has been
proposed in [45] along with the results of data analysis. Accord-
ing to the proposed protocol, the NIBP simulator is controlled
by a personal computer which is also linked to the OBMPD
under test [45], [46]. The evaluation protocol shown in Fig. 5
covers a range of simulated pressures, pulse rates, weak pul-
sations, and motion and tremor artifacts. The first part of the
protocol tests the monitor at six different pressures ranging
from low (60/30 mmHg) to high results (200/150 mmHg), with
five determinations at each target pressure. Then, a ramped
pressure sequence is applied to the monitor to test its ability Fig. 6. Differences between (a) oscillations generated by the CuffLink NIBP
to follow changing pressure patterns [46]. The cuff pressure is simulator and (b) oscillations taken from a human upper arm [35].
sampled by the oscillometric waveform at the pulse rate, and
hence, the pulse rate affects the measurement of the pressure. quoted CuffLink simulates SBP and DBP at fractions 50%
Therefore, the protocol plans the recording of the pressures over and 67% of the MBP value, while the BP Pump from Bio-
a range of pulse rates from 40 to 200 beats per minute (bpm) Tek adopts 54% and 59% for SBP and DBP, respectively [9].
with the simulated pressure kept fixed at 120/80 (93) mmHg. Moreover, some NIBP simulators choose to have the same
Next, to test the ability of the NIBP monitor to cope with ratio throughout the entire pressure range, while others use
artifacts, a low-frequency pressure may be added to the oscillo- a dynamic ratio that changes based on pressure [47]. Several
metric waveform in various degrees of severity. Similar ranges factors, including the type of simulator, settings, age of the
of severity of mixed low- and high-frequency tremor artifact are instrument, software version, pneumatic setup, and module
added. In all cases, the simulator has set to generate a pressure configuration bytes, affect the NIBP simulator expected values
of 120/80 mmHg at a pulse rate of 80 bpm. Finally, the ability [47]. Moreover, most commercial simulators generate artificial
of a monitor to cope with weak pulsations is tested by reducing waveforms presenting fixed smoothness and shape independent
the amplitude of the oscillometric waveform down to 10% of from systole and diastole phases [48]. Fig. 6 shows a qualitative
the nominal typical amplitude. example of the differences between an oscillogram generated
Of course, as any agreement among manufacturers is miss- by a commercial NIBP simulator (the CuffLink from Fluke
ing, NIBP simulators from different manufacturers usually give Biomedical) [48] and the oscillogram taken from a human
different results with the same settings. This is primarily due upper arm [35].
to a number of algorithm differences between manufacturers, For the aforementioned reasons, NIBP simulators cannot be
as for the OBPMDs [47]. Differences in NIBP simulator oscil- used as a substitute for clinical validations to measure the
lation amplitude ratios can be found. For example, the earlier accuracy of NIBP measurement devices, but as their output
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2399

is very stable [42], they are able to assess the measurement TABLE III
NIBP S IMULATOR T EST R ESULTS FOR H YPERTENSIVE S ETTING [47].
repeatability and reproducibility. SYS I S THE S YSTOLIC B LOOD P RESSURE , DIA I S THE D IASTOLIC
Basic procedures for evaluating an NIBP simulator assess- B LOOD P RESSURE , M EAS . MAP I S THE M EASURED M EAN A RTERIAL
ing its suitability and quality have been proposed in [42]. P RESSURE , C AL . MAP I S THE C ALCULATED M EAN A RTERIAL
P RESSURE , AND HR I S THE H EART R ATE
In static conditions, in fact, NIBP simulator performance is
almost always sufficient to provide a reference for BP mea-
surements. On the other hand, in dynamic conditions, there
are differences among different simulators. Therefore, both a
static calibration and a dynamic evaluation are proposed for
the NIBP simulator testing in [42]. In particular, in order to
evaluate a simulator, the process is divided into two parts. The
first part is evaluation under static conditions, which includes
static calibration, repeatability, and reproducibility tests of its
transducer. The second part includes dynamic testing of the
simulator to determine the stability of simulator output and
qualitatively assess the stability of the generated oscillometric
envelope shape, dependence of the envelope from the generated
pressure magnitude and heart rate, etc. [42].
However, since commercial simulators generate specific
shapes of pressure signals, a simulator output cannot always
be a good enough reference for the calibration. NIBP mon-
itors can only be calibrated (in terms of determining NIBP
monitor accuracy) using a certain, suitable, and prescribed
simulator [42].
In [47], a protocol to test the performance limits of five
NIBP simulators of different brands, most commonly used by
NIBP developers, medical monitoring companies, and biomed-
ical engineering departments throughout the world, by using
ten Advantage Model 2 OEM NIBP modules produced by
Suntech Medical, is presented. Performance limits have been
provided for both measured and calculated MBPs for four
preset simulations: hypotensive, normotensive, hypertensive,
and a neonatal setting. Each simulator setting was performed
three times per module for a total of 30 simulations per setting.
Statistical analysis was done by calculating the mean (average)
and standard deviation of SBP, DBP, and MBP for each group
of 30 simulations. The mean shows the shift from the target
value, and the standard deviation was used to determine the
span between the minimum and maximum limits. Low-quality
NIBP simulators have wider hypertension limits, and high-
quality NIBP simulators have tighter limits [47]. In Table III,
test results for hypertensive setting are shown. According to the
results, the authors of the paper recommend Fluke Biomedical
Cufflink and the Clinical Dynamics SmartArm. The Bio-Tek
BP Pump is also considered an acceptable instrument, but is
no longer commercially available, while the Fluke Biomedical
BP Pump 2 and the Metron QA-1290 are not recommended
since they present, respectively, maximum standard deviations
of 3.72 and 5.04 for the SBP [47].

C. Physiological Waveform Modeling and Simulators

Physiological signals are the most complex in nature and so
very difficult to be realistically reproduced. Arterial BP signal
has been analyzed by using different mathematical tools, such
as Fourier transform, in [49]. The proposed analysis methods
are focused to discover some properties in the frequency do-
main of the arterial BP signal and to perform the following:
1) to investigate its variability; 2) to extract features of the
2400 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

Fig. 7. (Top) Human signal, Morlet wavelet, SBP = 83 mmHg, DBP = 58 mmHg, HR = 74 bpm, and CuffLink signal. (Bottom) Morlet wavelet, SBP =
80 mmHg, DBP = 50 mmHg, and HR = 80 bpm [51].

pulses at certain pressure levels; and 3) to form BP profiles, forms (built into the machine) [51]. In order to investigate
studied by fuzzy sets determined by neural networks [49]. A such difference, in [51], five oscillometric waveforms have been
lot of research work deals with the application of the wavelet captured from real patients with a variety of measured BPs and
transform to the arterial BP signal analysis [49], [50], since compared with five oscillometric waveforms produced with a
wavelets have finite duration (compact support) as contrasted CuffLink simulator. In each case, the CuffLink simulator has
with Fourier methods based on sinusoids of infinite duration been configured in order to achieve the best match between the
[50]. Owing to the wavelet transform, it has also been possible simulated and the measured waveform [51]. Then, the analysis
to discern between human and simulated oscillometric wave- using continuous wavelet transform has been performed us-
forms to understand better the limitations found in current BP ing MATLAB wavelet toolbox. In Fig. 7, an example of the
simulators [51]. achieved results is reported showing a pair of human/simulated
In particular, considering that the random fluctuations in signals. The authors affirm that the most noticeable differences
the human body’s physiology are not modeled and reproduced were observed with a Morlet mother wavelet, in the ranges of
by a simulator, the presence of these fluctuations can serve scale 41–53 as well as 17–33. By observing Fig. 7, in fact, the
as a means to differentiate between real and machine-made repetitive pattern of the machine-generated signal—nearly iden-
oscillometric waveforms. tical for each heart beat—is readily observable. On the human-
In fact, the signals contained within each pulse period of generated signal, instead, there are clear variations of the
a machine-made signal (between two successive oscillometric position of the transform peaks with respect to scale. This indi-
waveform peaks) are some combination of scaled stored wave- cates some beat-to-beat variation that the machine is not capable
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2401

Fig. 8. High-level block diagram of the physiological signal simulator in [54].

of generating. These variations are most prominent at the larger These are, in fact, amplified for the purpose of displaying
scaling factors, which are related to lower frequency compo- on the oscilloscope. If the true amplitudes for these signals
nents. These variations were present in all five human wave- were to be employed instead, it would create problems due to
forms tested but absent in all machine-generated signals [51]. their small values, compared with the system noise. A linear
Recently, it has been shown that certain important patterns approximation routine was created for each physiological signal
can be discovered by observing a few dominant harmonics of for the purpose of scaling the amplitudes, with the exception
the arterial BP waveforms, acquired using a catheter–transducer of the electrocardiogram. The user can vary the systolic and
pressure measuring system [49]. In particular, there are two diastolic pressures of the arterial BP in the ranges of 90–106
major harmonics at work with all patients and all data sets and 50–65 mmHg, respectively [54].
considered in the paper. These two harmonics represent a total Various BP simulators have been proposed to evaluate and
of about 92% of the total energy of the arterial BP signal and improve the performance of an automatic sphygmomanometer
have been verified by analysis using Matlab. The power density [55]. However, as shown before, the pressure pulse waveform
is distributed to the first and second harmonics in such a manner produced in the automatic sphygmomanometer differs from
that the system seems to keep the total percentage of energy in the actual pulse produced in the human body. In general, it
the first and second harmonics stable and constant. is very difficult to reproduce the very low BP waveforms
Through the years, several contributions dealing with physio- using a fluid flowing in low-pressure tubes such as blood
logical signal models and simulators, coming from the research, in the human vessels [55]. To overcome such limitation, an
have been developed. improved simulator has been proposed in [55], including an
Nonlinear models for generating realistic BP signals [52], artificial arm model capable of feeding appropriate fluids that
also providing the generation of 24-h versions of them [53], can generate the BP waveform to test the full range of the
have been proposed. In particular, model parameters are ini- automatic sphygmomanometer. The schematic diagram of the
tialized using realistic distributions that can be modified so BP simulator is shown in Fig. 9. First of all, the oil used to
that different seeds produce different but repeatable biomedical transmit the pressure wave is sent to a pressure tank using a
signals. Since each of the parameters of each model has a circulation pump, and then, pressure is applied to the oil using
very clear physiological meaning, the performance of signal a compressor. The pressurized oil flows from the tank into a
processing algorithms (such as how they respond to specific pipe through a proportional control valve. The pressure wave
changes or contaminants) can be tested on a variety of artificial in the pipe is produced by opening and closing of the valve.
biomedical time series and may aid the teaching of the under- The fluid flows also in the vessels of an artificial human arm
lying mechanisms [53]. Moreover, open source C, Matlab, and manufactured with silicon, which is connected in parallel with
Java codes for these models are available from Physionet. the pipe, and then, it flows through a thin tube connected to the
A smart physiological signal simulator that is capable of end of the pipe and is stored again in the tank. The pressure
generating and displaying five physiological signals (electro- of the fluid flowing in the artificial arm is controlled in order
cardiogram, arterial BP wave, pulmonary artery wave, airway to reproduce the human BP, by means of a closed loop using a
pressure signal, and the phonocardiogram) onto an oscilloscope pressure transducer to monitor the pressure in the pipe.
has been proposed in [54] for didactic aims. It does this by The work described in [56], funded by the project of “The
first generating the physiological signals on a microcontroller establishment of new measurement standards for demands”
by linearly approximating their amplitudes and then generating from the Korean Research Institute of Standards and Science
the signals in real time for the oscilloscope using a high- (KRISS), developed a BP simulator for the monitoring of OBP-
speed multiplexing hardware interface. Fig. 8 shows the high- MDs and compared this KRISS simulator with commercial
level block diagram of the simulator. The microcontroller user simulators. The simulator consisted of a stepping motor, a lever
interface consists of standalone switches that are used as an assembly, a bellows, and a pressure transducer. To check the
alternative to control the simulator, and the microcontroller-to- accuracy of the proposed system, a pressure gauge was used
oscilloscope interface is the electronics used to produce phys- as a reference (Digiquartz/model No.6000-15G, Paroscientific
iological signals for displaying purposes. A series of lookup Inc.) having an error below 0.01% and traceable to the national
tables is used to generate the physiological signals that the user standard of Korea by the authors themselves. The measured
will view on their PC or oscilloscope. Each lookup table is signal was digitized (model No. 6671-003, Paroscientific Inc.)
200 bytes long and describes the digital equivalent of the analog and recorded in a notebook computer. The data acquisition
amplitude representation for each of the physiological signals. was controlled by a software written in LabView. From the
2402 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

Fig. 9. Schematic diagram of the BP simulator in [55].

results, authors concluded that the simulator could work accu- Digital recordings are ideal for these purposes [57]. By making
rately for pressure generation because the simulator showed a well-characterized clinical data available to researchers, these
relative repeatability error below 0.1%, which was defined as databases make it possible to formulate and answer numerous
the ratio between the standard deviation and the average value physiological questions, without the necessity of developing a
of the generated pressure signals [56]. The KRISS simulator new set of reference data [57]. In 1999, researchers at Boston’s
was compared with three commercial simulators by controlling Beth Israel Deaconess Medical Center, Boston University,
each device to generate a systolic pressure of 120 mmHg, a McGill University, and MIT inaugurated Physionet, a Web site
diastolic pressure of 80 mmHg, and a pulse rate of 75 beats for the biomedical research community, under the auspices
per second. For this comparison, a commercial OBPMD (model of the National Center for Research Resources of the U.S.
770A, Omron Co.) was connected to the four simulators under National Institutes of Health [58]. PhysioBank, one of the three
test, and each value of BP was recorded five times for each Web site components, is a database collection of a variety of
simulator. The results showed that each simulator generated digitized physiological signals. It includes the Multi-parameter
different values in spite of the same conditions and same Intelligent Monitoring for Intensive Care II database, providing
OBPMD. The authors suppose this difference due to the dif- nursing data from over 30 000 patients, together with high-
ference both in accuracy of pressure sensors adopted in each resolution waveform data (mainly electrocardiogram and
simulator and in the algorithms used in each simulator for signal arterial BP) from over 2700 of these patients. Unfortunately,
generation [56]. oscillometric BP waveforms are not included in the collection
Finally, the measurement repeatability of six models of com- yet [23].
mercial OBPMDs from five companies using the KRISS simu- A database of validated oscillograms is one of the objec-
lator was investigated, too. In particular, the KRISS simulator tives of the EU project “Simulator for NIBP” [35], [59]. This
was controlled to generate two conditions: 1) systolic pressure European Commission funded multicenter study is aimed at
of 120 mmHg—diastolic pressure of 80 mmHg—pulse of 75 developing an NIBP simulator that is able to reproduce realistic
and 2) systolic pressure of 150 mmHg—diastolic pressure of BPs, based on a library of prerecorded waveforms from a range
94 mmHg—pulse of 75. The measurement was then conducted of patient groups. This digital database shall represent typical
five times for each condition. For each measurement, the max- oscillations taken for representative groups of subjects, due to
imum pressure difference was 8 mmHg for condition 1). The age, sex, disease, measuring conditions (rest, movement, venti-
authors suppose this difference due to the different algorithms lated, etc.), and artifacts (from the patient or the environment)
of BP simulators adopted by each manufacturer to calibrate [59]. At 2006, more than 1000 signals from more than 600
each OBPMD. This hypothesis seems to be borne out by similar subject were gathered [35].
results shown by two models from the same company [56]. The concept of oscillometric pulsation waveform database,
The authors expect to develop a BP simulator for Korean by allowing the test of various algorithms aiming to characterize
the collection of clinical data from the Koreans [56]. the oscillometric BP waveforms in order to distinguish between
healthy people and people with cardiovascular problems (ar-
teriosclerosis, etc.), has been introduced in [31]. The concept
D. Physiological Waveform Databases
is based on oscillometric data retrieving during cuff deflation
Researchers, developers, and evaluators of algorithms and and on reference BP measurements by auscultation. The data
systems for the physiological data analysis require physiolog- which are collected about each of the considered patients are,
ical signal databases as essential resource [57]. Testing proce- e.g., date and time of measurement, age, gender, height, weight,
dures, in fact, need to operate on realistic data and to satisfy diagnosis and medication of the measured person, auscultation
repeatability and reproducibility requirements to establish if an and oscillometric values of BP, etc. Together with the data,
observed difference in the results can be due to a difference in oscillometric pulsations and cuff pressure are saved into the
the objects under test or to a difference in the input data. database [31].
BALESTRIERI AND RAPUANO: INSTRUMENTS AND METHODS FOR CALIBRATION OF OBPMDs 2403

When dealing with digital databases, the problem of the R EFERENCES

used electronic format arises. Some legacy programs have been [1] G. Paratia, A. Fainia, and P. Castiglioni, “Accuracy of blood pressure mea-
developed to digitize biomedical signals into electronic records, surement: Sphygmomanometer calibration and beyond,” J. Hypertens.,
usually restricted to a single format [60]. A possible alternative vol. 24, no. 10, pp. 1915–1918, Oct. 2006.
[2] M. J. Turner, C. Speechly, and N. Bignell, “Sphygmomanometer
is a program for improving biosignal databases capable of calibration—Why, how and how often?” Aust. Fam. Physician, vol. 36,
recognizing and digitizing any kind of biosignal paper records, no. 10, pp. 834–837, Oct. 2007.
grids, large strips, and different speeds by means of an easy [3] IEEE Standard for Objective Measurement of Systemic Arterial Blood
Pressure in Humans, IEEE P1721.
procedure as proposed in [60]. [4] E. Balestrieri and S. Rapuano, “Standard calibration procedures for auto-
mated non-invasive measurement of blood pressure,” Int. J. Adv. Media
E. Discussion Commun., vol. 3, no. 1, pp. 226–246, Jun. 2009.
[5] E. Balestrieri, P. Daponte, and S. Rapuano, “Open questions on unified
To be a good measurement reference standard, an NIBP approach for calibration of oscillometric blood pressure measurement
devices,” in Proc. MeMeA—Int. Workshop Med. Meas. Appl. Cetraro,
simulator must be suitably calibrated itself to assure the re- Italy, May 29–30, 2009, pp. 206–211.
quired metrological characteristics in terms of accuracy, pre- [6] R. Khandpur, Biomedical Instrumentation: Technology and Applications.
cision, repeatability, reproducibility, and traceability. Currently, New York: McGraw-Hill Professional, Nov. 2004.
[7] S. Rithalia, M. Sun, and R. Jones, Blood Pressure Measurement. Boca
NIBP simulators have shown to be only stable enough to test Raton, FL: CRC Press, 2000.
repeatability and reproducibility of NIBP monitors [41], [42]. [8] T. K. Kim, Y. J. Chee, J. S. Lee, S. W. Nam, and I. Y. Kim, “A new blood
However, an agreement concerning NIBP simulator calibration pressure measurement using dual-cuffs,” in Proc. Comput. Cardiol., 2008,
vol. 35, pp. 165–168.
does not exist yet, as well as a standard calibration procedure. [9] A. Ball-llovera, R. Del Rey, R. Ruso, J. Ramos, O. Batista, and I. Niubo,
Although some efforts have been done to evaluate and compare “An experience in implementing the oscillometric algorithm for the non-
NIBP simulator accuracy and repeatability [47], [56], their invasive determination of human blood pressure,” in Proc. 25th Annu. Int.
Conf. IEEE EMBS, Cancun, Mexico, Sep. 17–21, 2003, pp. 3173–3175.
exhaustive metrological characterization has not been carried [10] A. Sapinski, “Standard algorithm of blood-pressure measurement by the
out yet. oscillometric method,” Med. Biol. Eng. Comput., vol. 30, no. 6, p. 671,
A possible solution to overcome the problem of the differ- Nov. 1992, Letters to the Editor.
[11] W. Kaspari, “Blood pressure: Differential auscultatoly technique,” Med.
ences in NIBP simulator generated waveforms and the phys- Electron., Apr. 1995.
iological ones could be the creation of a public physiological [12] K. Yamakoshi and S. Tanka, “Standard algorithm of blood-pressure mea-
database from which the waveforms are taken to generate. This surement by the oscillometric method,” Med. Biol. Eng. Comput., vol. 31,
no. 2, p. 204, Mar. 1993, Letters to the Editor.
database, currently missing, could also be taken as a standard [13] L. A. Geddes, Handbook of Blood Pressure Measurement. Clifton, NJ:
reference to calibrate the NIBP simulators after a comprehen- Humana Press Inc., 1991.
sive metrological characterization of the physiological wave- [14] H. Sorvoja, R. Myllylä, P. Kärjä-Koskenkari, J. Koskenkari, M. Lilja, and
Y. A. Kesäniemi, “Accuracy comparison of oscillometric and electronic
form has been achieved. Although this topic has recently caught palpation blood pressure measuring methods using intra-arterial method
researchers’ interest [51], another open question arises concern- as a reference,” Molecular Quantum Acoust., vol. 26, pp. 235–260, 2005.
ing the way in which the reference physiological waveforms [15] J. Talts, Continuous Non-Invasive Blood Pressure Measurement: Compar-
ative and Methodological Studies of the Differential Servo-Oscillometric
have to be acquired and recorded. Some specialized systems Method. Tartu, Estonia: Tartu Univ. Press, 2004.
to acquire a database of oscillometric BP waveforms from a [16] G. Drziewiecki, “Noninvasive assessment of arterial blood pressure and
wrist cuff have been proposed [51], [61]. Unfortunately, there is mechanics,” in The Biomedical Engineering Handbook. Boca Raton,
FL: CRC Press, 1995, pp. 1196–1211.
currently no defined independent standard for wearable cuffless [17] A. Sapinski, “Theoretical basis for proposed standard algorithm of blood
BP measurement devices. Existing standards for evaluating pressure measurement by the sphygmooscillographic method,” J. Clin.
sphygmomanometers are only intended for devices that are Eng., vol. 22, no. 3, pp. 171–174, May 1997.
[18] G. A. van Montfrans, “Oscillometric blood pressure measurement:
used with an occluding cuff and therefore do not cover all Progress and problems,” Blood Press. Monit., vol. 6, no. 6, pp. 287–290,
aspects needed for the emerging wearable devices [62]. Dec. 2001.
[19] T. J. Brinton, B. Cotter, M. T. Kailasam, D. L. Brown, S. S. Chio,
D. T. O’Connor, and A. N. DeMaria, “Development and validation of a
VI. C ONCLUSION noninvasive method to determine arterial pressure and vascular compli-
ance,” Amer. J. Cardiol., vol. 80, no. 3, pp. 323–330, Aug. 1997.
An overview of the research and manufacturer proposals for [20] S. Narus, T. Egbert, T. K. Lee, J. Lu, and D. Westenskow, “Noninvasive
performing bench dynamic accuracy testing of NIBP devices blood pressure monitoring from the supraorbital artery using an artificial
relying on the oscillometric technique has been presented. In or- neural network oscillometric algorithm,” J. Clin. Monit., vol. 11, no. 5,
der to remove logistical difficulties, time-consuming processes, pp. 289–297, Sep. 1995.
[21] T. G. Pickering, “Principles and techniques of blood pressure measure-
and high costs involved in clinical validations, as well as to im- ment,” Cardiol. Clin., vol. 20, no. 2, pp. 207–223, May 2002.
prove the reproducibility of the calibration procedures, several [22] K. Ruiter, Optimizing Simulation Results With Your SimCube Oscillomet-
simulators have been developed for NIBP device calibration. ric NIBP Simulator. Sun Valley, CA: Pronk Technol. Inc., 2004.
[23] J. Jilek and M. Stork, “Bench testing of oscillometric blood pressure
Unfortunately, at the present time, conventional simulators are monitors,” in Proc. Appl. Electron., 2006, pp. 67–70.
not able to assess the OBPMD accuracy. Nevertheless, re- [24] A. J. Sims, C. A. Reay, D. R. Bousfield, J. A. Menes, and A. Murray,
search work seems to support the future replacement of clinical “Low-cost oscillometric non-invasive blood pressure monitors: Device
repeatability and device differences,” Physiol. Meas., vol. 26, no. 4,
study validation in favor of simulator evaluation. The “ideal” pp. 441–445, Aug. 2005.
simulator requires a comprehensive database of physiologi- [25] A. Stang, S. Moebus, S. Mohlenkamp, N. Dragano, A. Schmermund,
cal waveforms, a validation protocol that assesses accuracy E. M. Beck, J. Siegrist, R. Erbel, and K. H. Jockel, “Algorithms for
converting random-zero to automated oscillometric blood pressure val-
over a comprehensive range of conditions and its independent ues, and vice versa,” Amer. J. Epidemiol., vol. 164, no. 1, pp. 85–94,
verification. Jul. 2006.
2404 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 9, SEPTEMBER 2010

[26] C. T. Lin, S. H. Liu, J. J. Wang, and Z. C. Wen, “Reduction of interference [50] R. G. Hohlfeld, C. Rajagopalan, and G. W. Neff, Wavelet Signal Process-
in oscillometric arterial blood pressure measurement using fuzzy logic,” ing of Physiologic Waveforms. Attleboro, MA: Wavelet Technol. Inc.,
IEEE Trans. Biomed. Eng., vol. 50, no. 4, pp. 432–441, Apr. 2003. 2003.
[27] D. Zheng and A. Murray, “Estimation of mean blood pressure from oscil- [51] N. Rank, V. Groza, and R. Leca, “Characterization of blood pressure
lometric and manual methods,” in Proc. Comput. Cardiol., 2008, vol. 35, oscillometric waveforms,” in Proc. 2nd Int. Conf. Syst. Netw. Commun.,
pp. 941–944. Cap Esterel, France, Aug. 2007, pp. 79–82.
[28] H. S. Choi, H. S. Myoung, H. K. Lee, H. D. Park, and K. J. Lee, “A new [52] G. D. Clifford and P. E. McSharry, “A nonlinear artificial model for gen-
noise reduction method for oscillometric blood pressure measurement,” erating realistic correlated ECG, BP and respiration,” in Proc. Biosignal,
in Proc. 30th Annu. Int. IEEE EMBS Conf., Vancouver, BC, Canada, Aug. Brno, Czech Republic, Jun. 2004, pp. 358–360.
20–24, 2008, pp. 270–272. [53] G. D. Clifford and P. E. McSharry, “Generating 24-Hour ECG, BP and
[29] V. Jazbinsek, J. Luznik, and Z. Trontelj, “Influence of different representa- respiratory signals with realistic linear and nonlinear clinical character-
tions of the oscillometric index on automatic determination of the systolic istics using a nonlinear model,” in Proc. Comput. Cardiol., Chicago, IL,
and diastolic blood pressures,” in Proc. IFMBE, 2008, pp. 216–220. Sep. 2004, pp. 709–712.
[30] H. D. Kiers, J. M. Hofstra, and J. F. M. Wetzels, “Oscillometric blood pres- [54] R. Phan and R. Leung, “Smart physiological signal simulator,” in Proc.
sure measurements: Differences between measured and calculated mean 5th Int. Conf. Upcoming Eng., Waterloo, ON, Canada, May 2006.
arterial pressure,” Neth. J. Med., vol. 66, no. 11, pp. 474–479, Dec. 2008. [55] C. H. Kim, G. B. Han, H. C. Lee, Y. J. Kim, K. G. Nam, G. S. Gong,
[31] V. Fabián, M. Janouch, L. Nováková, and O. Štepánková, “Compara- Y. J. Lee, K. S. Lee, G. R. Jeon, and S. Y. Ye, “Blood pressure simulator
tive study of non-invasive blood pressure measurement methods in el- using an optimal controller with disturbance observer,” Int. J. Control,
derly people,” in Proc. IEEE EMBS Cité Internationale, Lyon, France, Autom., Syst., vol. 5, no. 6, pp. 643–651, 2007.
Aug. 23–26, 2007, pp. 612–615. [56] H. W. Song, S. J. Lee, Y. K. Park, I. M. Choi, and S. Y. Woo, “The
[32] H. D. Park, H. S. Choi, S. P. Cho, and K. J. Lee, “Minimization of development of a blood pressure simulator in KRISS,” in Proc. 30th
artifact using adaptive digital filter during the oscillometric blood pressure Annu. Int. IEEE EMBS Conf., Vancouver, BC, Canada, Aug. 2008,
measurement,” in Proc. Int. Conf. Control, Autom. Syst., Seoul, Korea, pp. 3285–3288.
[57] G. B. Moody and R. G. Mark, “A database to support development and
Oct. 17–20, 2007, pp. 612–615.
evaluation of intelligence intensive care monitoring,” in Proc. Comput.
[33] B. A. Friedman, “Assessment of the validation of blood pressure monitors:
Cardiol., Indianapolis, IN, Sep. 1996, pp. 657–660.
A statistical reappraisal,” Blood Press. Monit., vol. 13, no. 4, pp. 187–191,
[58] G. B. Moody, R. G. Mark, and A. L. Goldberger, “PhysioNet: A Web-
Aug. 2008.
based resource for the study of physiologic signals,” IEEE Eng. Med. Biol.
[34] W. Gerina, A. R. Schwartz, J. E. Schwartz, T. G. Pickering, Mag., vol. 20, no. 3, pp. 70–75, May/Jun. 2001.
K. W. Davidson, J. Bress, E. O’Brien, and N. Atkins, “Limitations of cur- [59] [Online]: Available: http://cordis.europa.eu
rent validation protocols for home blood pressure monitors for individual [60] J. Millet-Roig, A. Mocholi-Salcedo, J. J. Lopez-Soriano, R. Garcia-
patients,” Blood Press. Monit., vol. 7, no. 6, pp. 313–318, Dec. 2002. Civera, and R. Ruiz-Granell, “A method capable of digitising every type
[35] APEC/APLMF Training Courses in Legal Metrology, Handbook on Au- of biosignals paper recordings as a tool for database improving,” in Proc.
tomated Sphygmomanometers, 2006. 20th Annu. Int. Conf. IEEE Eng. Med. Biol. Soc., Hong Kong, Oct./Nov.
[36] J. N. Amoore, Y. Lemesre, I. C. Murray, E. Vacher, S. Mieke, S. T. King, 1998, pp. 1238–1241.
F. E. Smith, and A. Murray, “Validation of oscillometric non-invasive [61] J. Jilek and M. Stork, “Oscillometric pressure pulse waveforms:
blood pressure measurement devices using simulators,” Blood Press. Their current and prospective applications in biomedical instrumenta-
Monit., vol. 12, no. 4, pp. 251–253, 2007. tion,” in Proc. 13th WSEAS Int. Conf. Syst., Rodos, Greece, Jul. 2009,
[37] J. N. Amoore, E. Vacher, I. C. Murray, S. Mieke, S. T. King, pp. 133–138.
F. E. Smith, and A. Murray, “Can a simulator that regenerates physi- [62] X. Y. Xiang, C. C. Y. Poon, and Y.-T. Zhang, “Modeling of the cuffless
ological waveforms evaluate oscillometric non-invasive blood pressure blood pressure measurement errors for the evaluation of a wearable med-
devices?” Blood Press. Monit., vol. 11, no. 2, pp. 63–67, Apr. 2006. ical device,” in Proc. 3rd IEEE-EMBS Int. Summer School Symp. Med.
[38] JCGM 200:2008, “international vocabulary of metrology—basic and Devices Biosensors, 2006, pp. 105–108.
general concepts and associated terms,” 3rd ed., VIM, 2008.
[39] V. Hurlbut, “NIST calibration and conformance testing. What’s it all
about?” Nova Scotia Provincial Blood Coordinating Program Presenta- Eulalia Balestrieri received the M.S. degree in soft-
tion, Mar. 2005. ware engineering, with a thesis dealing with an im-
[40] CAS Medical Systems Inc., “performance of the CAS oscillometric al- age processing-based method for stress classification
gorithm when compared against various commercially available NIBP in digital telecommunication networks, and the Ph.D.
simulators,” Branford, CT, White paper, Oct. 2007. degree in information engineering from the Univer-
[41] G. Gersak and J. Drnovsek, “Evaluation of non-invasive blood pressure sity of Sannio, Benevento, Italy, in 2003 and 2007,
simulators,” in Proc. Mediterranean Conf. Med. Biomed. Eng. Comput., respectively.
Ljubljana, Slovenia, Jun. 2007, pp. 342–345. Then, she joined the research activities carried out
[42] G. Gersak, A. Zemva, and J. Drnovsek, “A procedure for evaluation at the Laboratory of Signal Processing and Mea-
of non-invasive blood pressure simulators,” Med. Biol. Eng. Comput., surement Information, University of Sannio. Her
vol. 47, no. 12, pp. 1221–1228, Dec. 2009. research interests include digital signal processing
[43] K. Ng and C. F. Small, “Design overview of a microcomputer-controlled for measurement in telecommunications, data converter characterization, and
NIBP simulator for evaluation of non-invasive blood pressure monitors,” medical measurements.
in Proc. Comput. Cardiol., Durham, NC, Oct. 1992, pp. 551–554.
[44] P. D. Davis, J. L. Dennis, and R. Railton, “Evaluation of the A&D UA-767
and Welch Allyn Spot Vital Signs noninvasive blood pressure monitors Sergio Rapuano (M’00–SM’10) received the M.S.
using a blood pressure simulator,” J. Hum. Hypertens., vol. 19, no. 3, degree (with honors) in electronic engineering and
pp. 197–203, Mar. 2005. the Ph.D. degree in computer science, telecommu-
[45] W. B. Geake, J. N. Amoore, and D. H. T. Scott, “An automated system nications, and applied electromagnetism from the
for the functional evaluation of oscillometric non-invasive blood pressure University of Salerno, Salerno, Italy.
monitors,” J. Med. Eng. Technol., vol. 19, no. 5, pp. 162–176, Sep. 1995. Since 2002, he has been an Assistant Professor of
[46] J. N. Amoore and W. B. Geake, “Evaluation of the Critikon 8100 and electric and electronic measurement with the Faculty
Spacelabs 90207 non-invasive blood pressure monitors using a test simu- of Engineering, University of Sannio, Benevento,
lator,” J. Hum. Hypertens., vol. 11, no. 3, pp. 163–169, Mar. 1997. Italy. His research interests include digital signal
[47] SunTech Medical Application Note 82-0053-00 Rev., “ANIBP simulator processing for measurement in telecommunications,
limits with the advantage OEM NIBP module series,” Eynsham, U.K., data-converter characterization, distributed measure-
2000. ment systems, virtual laboratories, and medical measurement.
[48] Fluke Biomedical, “CuffLink non-invasive blood pressure simulator,” Dr. Rapuano is a member of the IEEE Instrumentation and Measurement
Everett, WA, 2007. [Online]. Available. http://news.fluke.com/busen/ Society TC-10, Secretary of the TC-23 Working Group on e-tools for Education
home/default.htm in Instrumentation and Measurement, and the Cochairman/Official Reporter
[49] J. Ebenal, S. Vasana, C. Clinton, D. Cox, and T. Shine, “Arterial blood of the TC-25 Subcommittee on Objective Blood Pressure Measurement. He
pressure system modeling and signal analysis,” in Proc. IEEE Int. Symp. was the recipient of the IEEE Instrumentation and Measurement Society
Comput. Intell. Robot. Autom., Jacksonville, FL, Jun. 2007, pp. 386–391. Outstanding Young Engineer Award in 2007.