Augmented blood pressure measurement through the noninvasive estimation of physiological arterial pressure variability

Fifteen adults (seven male and eight female subjects, age ranging from 19 to 50 years) with no known history of heart disease participated in this study. All subjects provided informed consent prior to the measurement, in accordance with the guidelines of the Institutional Research Ethics Board.

We developed a BP measurement unit that includes two pressure cuffs, each connected to an analog pressure transducer. An illustrative block diagram is shown in figure 1. The Vernier pressure transducer (SenSym SDX05D4, Beaverton, OR, USA) and National Instruments^TM (Austin, TX, USA) hardware and software are used to acquire the arterial pulse wave data from the two arms. The Vernier pressure transducer operates on a dc supply voltage of 5 V and converts the mechanical vibrations exerted by the blood circulation on the cuff to an analog output voltage signal in the range of 0–5 V. The conversion factor is 56.11 mmHg V⁻¹.

Zoom In Zoom Out Reset image size

A first BP cuff is placed on the subject's upper left arm and is inflated to a constant pressure, using a manual pump, and is employed to retrieve the continuous arterial pulse waveform. A second BP cuff is placed on the upper right arm and controlled by an automatic 6 V dc mini air pump. The pump operates with a pushbutton mounted on the prototype; once pressed, the button turns the pump on and the brachial cuff is gradually inflated and then deflated to conduct the oscillometric measurement. A screw-controlled manual pressure release valve is connected in-line with the brachial cuff. The rotation of the screw determines the deflation rate of the cuff (Ahmad et al 2010).

The analog voltage outputs of the two transducers (corresponding to the two BP cuffs) are fed to two simultaneously sampled analog channels of the National Instruments^TM C Series 9239 analog input module (NI-9239) mounted on a CompactDAQ data acquisition system. The analog signals are conditioned, buffered and then sampled, at a frequency (f_s) of 1.613 kHz, by a 24 bit delta-sigma analog-to-digital converter. Prior to digitization, the signal passes through an in-built anti-aliasing filter on the data acquisition board with a cutoff frequency equal to 0.453f_s. The quantized signals are then transmitted to a personal computer via a universal serial bus (USB) cable. The voltage supply to the Vernier pressure transducers is provided from a National Instruments^TM C Series 9263 4-Channel, 16 bit, ± 10 V analog output module (NI-9263) also mounted on the CompactDAQ system.

National Instruments^TM LabVIEW development environment is used for data acquisition of the constant pressure arterial waveform and the oscillometric BP signals. We developed a customized LabVIEW user interface that displays the acquired arterial pulse waveform and oscillometric signal data in real-time. Signal processing and statistical analysis are performed offline using Matlab® (The MathWorks Inc., Natick, MA, USA). To minimize the effects of 60 Hz contamination and other interferers, the signals were digitally filtered using a third-order FIR low-pass filter with a cutoff frequency of 30 Hz prior to further analysis.

The subject is comfortably seated and asked to relax with minimum movement. The first cuff is placed on the upper left arm for the constant cuff pressure arterial pulse waveform measurement, the second cuff is placed on the upper right arm for oscillometric inflation and an automated BP device (UFIT, Biosign Inc., Toronto, Canada) is placed on the wrist of the right arm. UFIT is a commercial device that uses a wrist cuff to estimate the systolic and diastolic BP. In a study on 87 participants, the manufacturer reports an error (mean ± SD) of 0.3 ± 6.9 mmHg for the systolic pressure and 0.2 ± 5.2 mmHg for the diastolic pressure, compared to measurements with two trained observers (Biosign report 2011), and so the device meets the performance requirements of the AAMI/ANSI-SP10 (2002) standard for automated sphygmomanometers. The UFIT device is used to find the systolic and diastolic values at the start of every recording, but its cuff is kept deflated for the remainder of the session, and does not play a further role in the measurement.

The cuff on the left arm is inflated to reach a constant pressure, close to the diastolic pressure of the subject determined by the UFIT device (DBP + 3 mmHg), which was the minimal pressure allowing a clear visual identification of the pulse while at the same time minimizing any discomfort. This applied pressure value was determined experimentally, and the detection algorithm was able to detect pulses while the BP varied throughout the whole measurement. Once the pressure in the cuff on the left arm reaches a constant value, an oscillometric measurement is made on the right arm. In this arm, the cuff is inflated to a pressure higher than the systolic pressure, determined by the UFIT device (SBP + 5 mmHg), and then slowly deflated. Acquisition of data from the two cuffs is triggered simultaneously by the control software, and lasts 90 s, with the constant pressure cuff providing the continuous arterial pulse waveform and the oscillometric cuff generating the SBP and DBP values. The appropriate interval over which the mean SBP or DBP should be estimated is not known; however, a duration in the range of 1–2 min is expected to include the major short-term physiological oscillations (Elghozi 2008).

A single experimental session consists of a reference reading using UFIT followed by a BP recording from the two arms (oscillometric and constant pressure), in parallel, with these two steps repeated five times. At the end of the session, an additional recording is performed on both arms in which the subject is asked to breathe deeply, at a rate of 8 s/breath (i.e. breathing frequency of 0.125 Hz). A software-generated metronome is set up on the monitor, during this recording, to help the subject maintain the required breathing frequency. Deep breathing is employed because it typically results in particularly large excursions in the arterial pulse waveform.

A block diagram of the proposed approach is shown in figure A1. A customized peak detection algorithm is applied to the two recorded signals in order to detect the peaks (systolic points) and troughs (diastolic points) of the pulse waveform. The peak detection algorithm is based on an existing method for ECG peak detection (Kohler et al 2003) with minor alterations compensating for the shape disparity between an ECG pulse and a BP pulse (Chen 2010). In the first iteration, the continuous pulse waveform (shown in figure 2) is filtered using a third-order IIR Butterworth band-pass filter applied with cutoffs at 0.25 and 2.5 Hz, with the signal fed forward and then backward. The upward and downward zero crossings are then detected in order to find the start and end of each pulse, and the peaks/troughs are identified. A second iteration uses the detected peaks to build a band-pass filter, centered at the heart rate frequency, calculated as the mean of the inter-peak intervals of the first iteration. The band-pass filter has cutoff frequencies at twice the standard deviation of the heart rate on each side, or HR − 2 × std_HR and HR + 2 × std_HR, with 'HR' being the heart rate and 'std HR' its standard deviation. The zero crossings are identified during the second iteration and the final peaks and troughs represent, once again, the maximum and minimum values in between the upward and downward crossings, respectively (Chen 2010).

Zoom In Zoom Out Reset image size

A second-order high-pass Butterworth filter is applied to the OMW (shown in figure 3), with a cutoff frequency of 0.1 Hz. The filter suppresses the frequency components that are related to the slowly deflating cuff pressure and allows the remaining components of the signal to pass through (Geddes et al 1982, Sorvoja et al 2005). These components correspond to the subject's pressure pulses that form the desired OMW.

Zoom In Zoom Out Reset image size

Once the OMW is extracted, the oscillometric pulse index (OPI) is defined as the area obtained by integrating each pulse (Jazbinsek et al 2005, Ball-llovera 2003, Gershak et al 2006). The OPI values are used to construct a signal envelope that is interpolated and resampled at 100 Hz (Chen 2010).

The maximum amplitude algorithm (MAA), which has been reported to be the most popular algorithm for determining BP from OMWs (Baker et al 1997), is used to estimate the SBP and DBP values. MAA requires, as a first step, finding the point on the envelope that corresponds to the mean arterial pressure (MAP), which is estimated to be the maximum point on the OPI. Locating this value on the deflation curve allows us to obtain the MAP value in mmHg. Once plotted on the OPI, the MAP virtually divides the envelope into two parts: the left part represents the systolic side and the right part becomes the diastolic side. The MAA algorithm then utilizes systolic and diastolic characteristic ratios, determined empirically in large-scale studies, to find the points that correspond to SBP and DBP. In the literature, the systolic ratio ranges from 0.45 to 0.73 and the diastolic ratio ranges from 0.69 to 0.83 depending on specifics of the applied MAA. The height (amplitude A) of the maximum point is multiplied by the systolic ratio (r_s) and the diastolic ratio (r_D) in order to obtain the amplitude of the systolic point (located on the systolic side) and the diastolic point (located on the diastolic side), respectively (equation (1)). The ratios used in this study are 0.55 and 0.8 for r_s and r_D, respectively, based on a previous study by Chen (2010):

$$\begin{equation} {\rm SBP} = A \times r_s ,\quad {\rm DBP} = A \times r_D . \end{equation} \tag{ 1 }$$

Once the diastolic and systolic points are found on the envelope, they are mapped back to the deflation curve giving the SBP and DBP values in mmHg. The oscillometric algorithm explained in this section and used as the starting point of the proposed approach has been previously validated in our laboratory (Chen 2010).

In the conventional oscillometric measurement, the systolic BP corresponds to a specific heartbeat: the beat at which the BP became higher than the deflating cuff pressure and the blood was capable of flowing in the brachial artery. This specific moment is accompanied with a well-identified Korotkoff sound that can be heard by trained observers using a stethoscope. Similarly, the diastolic BP also corresponds to a specific beat. This correspondence is the basis of the time mapping explained in this section.

Following the execution of the peak detection algorithm, the estimated values of SBP and DBP on the interpolated envelope of the OMW are matched to the SBP and DBP beats in the continuous pulse waveform. This is done by finding the closest peak and trough to those values, respectively; the peak that is the closest to the SBP time point on the envelope is tagged as the SBP beat and the closest trough to the DBP time point on the envelope is tagged as the DBP beat. These points are then mapped to their respective corresponding beats on the constant pressure pulse waveform (figure 4).

Zoom In Zoom Out Reset image size

The final stage of the mapping consists of the calibration of the constant pressure pulse waveform, which is originally centered around the value of the external pressure applied by the cuff on the subject's arm. The calibration follows the equation

$$\begin{eqnarray} &&\fl y_{{\rm calibrated}} = \left[ {\frac{{{\rm OMW}_{{\rm SBP}} - {\rm OMW}_{{\rm DBP}} }}{{{\rm uncalibrated}_{{\rm SBP}} - {\rm uncalibrated}_{{\rm DBP}} }} \times (y_{{\rm uncalibrated}} - {\rm uncalibrated}_{{\rm DBP}} )} \right]\nonumber\\ &&+\, {\rm uncalibrated}_{{\rm DBP}} , \end{eqnarray} \tag{ 2 }$$

where OMW_SBP and OMW_DBP are the calibrated SBP and DBP values generated from the oscillometric method, uncalibrated_SBP and uncalibrated_DBP are the uncalibrated corresponding SBP and DBP points on the constant cuff pressure recording, and y_uncalibrated and y_calibrated (both functions of the time variable 't') indicate a given point in the constant cuff pressure curve before and after calibration, respectively. This equation results in a calibrated continuous recording from which we can obtain the calibrated SBP and DBP values for the duration of the measurement as well as the estimation of calibrated means, standard deviations, ranges and histograms. It should be noted that the generation of the statistical profile of the SBP and DBP during the measurement would not be affected by the location of the oscillometric estimates relative to the mean.

The next step is to identify the uncertainty in the SBP and DBP estimates that result from BP variability, and to classify outlier values. Descriptive statistics and histograms were obtained using SPSS (IBM, NY, USA). In the analysis of heart rate variability, short duration signals of less than 5 min are assumed to be at least approximately stationary (Task Force of the European Society of Cardiology and the North American Society of Pacing Electrophysiology 1996). As a result, we also assumed that the signals we analyzed were approximately stationary, since they are considerably shorter than 5 min.

The mean of the SBP and DBP values in the continuous BP waveform over the 90 s measurement interval (INT_MEAN_SBP and INT_MEAN_DBP) is calculated. The standard deviation around these means is also calculated (INT_STD_SBP and INT_STD_SBP). The SBP and DBP points estimated from the oscillometric measurement (i.e. OMW_SBP and OMW_DBP) are identified on the SBP and DBP histograms, respectively, in order to determine whether or not they are outliers. Locating the oscillometric values relative to the histogram could alert the user that the reading is potentially over- or underestimating their BP (represented by the mean over the measurement interval) and notifying them that a second reading may be needed to validate the first one. The three additional parameters for each of the SBP and DBP pressures (INT_MEAN, INT_STD and outlier indicator) are used to augment the conventional oscillometric BP measurement using information about BP variability over the measurement interval. Paired t-tests and Bland–Altman plots are used to compare the new measures and the conventional oscillometric SBP and DBP readings (i.e. OMW_SBP and OMW_DBP) and were performed using Analyse-it® (Leeds, UK).

Over the 90 measurements taken from 15 subjects (six readings per subject), the percentage of oscillometric SBP and DBP values (i.e. OMW_SBP and OMW_DBP) (1) located beyond ±5 mmHg from the mean values over the interval and (2) located at a distance exceeding the (2× SD) margin was evaluated in order to assess the contribution of BP variability to measurement uncertainty (Barnett, 1994). The first criterion corresponds to the maximum device accuracy error accepted by the SP10 standard (AAMI/ANSI-SP10 2002) and the second is used to determine outliers. It should be noted, however, that the proposed approach is not tied to any one criterion used for determining outliers. The criterion we suggest is one that is commonly used, but other criteria that may be less sensitive to the effects of the outliers on the sample mean and SD, such a criterion based on the upper/lower quartile ±1.5 times the interquartile range, could also be used (Wilcox 2001).

Since mapping is made between beats in pulse pressure waveforms obtained from the two arms, it is important that the variability in the two arms be highly correlated. Therefore, a subsidiary experiment was conducted on the 15 subjects, where a cuff was placed on both arms and inflated to the same constant pressure (diastolic BP + 3 mmHg) for 60 s. There was a correlation of 98.59 ± 1.41% (mean ± standard deviation, range 90.49–99.9%, over 15 samples) between the interpolated peak line signals (for SBP and DBP) acquired from both arms. This suggests that, at least in healthy subjects, the variability of the pulse waveform is synchronized and very similar in both arms (Eguchi et al 2007).