A Method for Localized Computation of Pulse Wave Velocity in Carotid Structure Ravindra B Patil, Krishnamoorthy P, and Shriram Sethuraman
Abstract— Pulse Wave Velocity (PWV) promises to be a useful clinical marker for noninvasive diagnosis of atherosclerosis. This work demonstrates the ability to perform localized carotid PWV measurements from the distention waveform derived from the Radio Frequency (RF) ultrasound signal using a carotid phantom setup. The proposed system consists of low cost custom-built ultrasound probe and algorithms for envelope detection, arterial wall identification, echo tracking, distension waveform computation and PWV estimation. The method is proposed on a phantom data acquired using custom-built prototype non-imaging probe. The proposed approach is non-image based and can be seamlessly integrated into existing clinical ultrasound scanners.
I. INTRODUCTION Atherosclerotic cardiovascular disease is one of the most common causes of mortality in the modern world. In most developed countries, about 10% of the population is affected by cardiovascular diseases. Also, in more than 90% of the reported cases of coronary heart disease (CHD), the problem is localized to the blood vessels (atherosclerotic disease) and not the heart [1]. Early detection and prevention of atherosclerosis is important to avoid fatalities due to cardiac events. Research studies [2] have indicated that measure of arterial stiffness is an important parameter for early detection of atherosclerosis. In clinical practice, pulse wave velocity (PWV) an advanced measurement technique is used to measure the arterial stiffness. Several studies [2] have shown that this parameter is an independent predictor of cardiovascular mortality in the elderly, hypertensive, diabetics, and patients with chronic renal failure as well as in the general population. It has been established as a highly reliable prognostic parameter for cardiovascular morbidity and mortality in varied populations [2].
arterial stiffness, the most widely used is aortic pulse wave velocity, specifically in the area running from the aortic arch or common carotid artery to the common femoral artery. Typically, the pulse wave is detected by pressure transducers or arterial tonometry. The measurement of carotid-femoral PWV is made by dividing the distance (from the carotid point to the femoral point) by the so-called transit time (the time of travel of the foot of the wave over the distance) [5]. Clearly, it is not possible to analyze the carotid and femoral waves simultaneously with a single probe. Therefore, the waveforms are obtained separately and then subsequently normalized with the electrocardiogram (ECG). A pulsed Doppler ultrasound with a linear array probe, synchronized with ECG and a two-second minimum sliding window is typically used. The examination begins with the patient in a supine position after locating the carotid artery with B-mode at the supraclavicular level (1-2 cm of the bifurcation). Further, the Doppler flow is identified simultaneously with ECG. The process is repeated on the common femoral artery in the groin. Each recording involves two or three cardiac cycles. To obtain the transit time, it is measured from the R wave of QRS to the foot of the waveform using digital calipers [6]. The drawbacks associated with the current approaches are:
The localized PWV of an artery has direct relationship with elastic modulus [3]. Recently, several methods have emerged for the estimation of PWV using ultrasound principles [4]. Among the different methods of evaluating ------------------------------------------------------------------------------------------Ravindra B Patil is with the Philips Research, Department of Health care Applications, Bangalore 560045 India, phone: 91-80-4189-1055; e-mail:
[email protected]. Krishnamoorthy P is with the Philips Research, Department of Health care Applications, Bangalore 560045 India; e-mail:
[email protected]. Shriram Sethuraman is with the Philips Research North America NY 10510 USA; e-mail:
[email protected].
978-1-4244-9270-1/15/$31.00 ©2015 IEEE
1898
1.
Measurement has to be obtained from two locations of anatomical structures with gated ECG and the measured value is approximate as the exact distance between carotid and femoral arteries varies between subjects
2.
The carotid-femoral method provides the measure of global stiffness and not of localized artery of interest
3.
Additional devices such as ECG needs to be interfaced to obtain the transit time
4.
It is cumbersome to measure the PWV due to different anatomical sites being considered and also the usability perspective of device is low
5.
Prolonged learning period in order to become an experienced observer and that the devices used lack versatility
6.
Technical difficulty in obtaining measurements and anatomical limitations of some patients make it necessary to find faster and more versatile methods for measuring PWV
In this work, we address a few limitations of the current state of the art. Specifically, we propose a system and method for automated local PWV estimation with a low cost Doppler ultrasound probe. The automated algorithm computes the arterial wall distention waveform from the raw ultrasound RF signals and therefore does not rely on the expertise of the operator to interpret images. In addition, this is a single probe approach aimed at estimating the local stiffness through PWV measurements and not subject to the errors of a global approach. A non-imaging approach lends itself to a low-cost solution. An important feature in the proposed approach is automated vessel localization that minimizes operator dependence in making the measurements. In this paper, we present results obtained from testing the system and method on a flow phantom. Specifically, we demonstrate the ability to reliably obtain distension waveforms from adjacent locations on a short vessel segment and further compute the PWV.
mL/min was pumped through the flow channel using a CompuFlow 1000 pump (Shelley Medical Imaging Technologies). The blood mimicking fluid has acoustic and physical properties similar to human blood and was formulated according to IEC 1685 standards. The experimental setup is shown in the Fig. 2.
II. MATERIALS AND METHODS The basic principle of proposed PWV measurement is illustrated in Fig. 1. Data Acquisition - Custom Probe - Vessel Localization
Figure 2. Phantom experimental setup and probe configuration
Signal Conditioning -Noise Removal -Envelope detection
A 4 MHz prototype Doppler ultrasound probe (element size 5 mm x 5 mm) was utilized to obtain RF data from the flow phantom. The RF data from 2 elements (2 A-lines) that are separated a fixed distance apart (36 mm in this case) is obtained. The ultrasound pulse-echo measurements were made using a custom-built system capable of obtaining RF and Doppler data. The ultrasound transmits and data acquisition parameters were controlled using a LabView interface.
Wall detection -Peak detection -Adaptive Threshold
PWV Estimation -Systolic Rise Identification -Computation of temporal shift
Distension Computation - Gating Mechanism - Cross correlation
Typical experimental parameters include: acquisition time of 5 s, maximum depth of 70 mm, pulse repetition frequency (PRF) of 5 KHz. The ultrasound echoes are digitized at a sampling rate of 20 MHz. B. Signal Conditioning
Figure 1. Flow diagram of the proposed algorithm The summary of the steps involved in estimation of localized PWV are described below: A. Data Acquisition An experimental setup consists of a commercially available flow phantom (Model 524, ATS Laboratories, Bridgeport, CT, USA) having a vessel channel mimicking a carotid artery measuring 6 mm in diameter. The vessel channel was approximately 2 cm from the surface of the phantom. Blood mimicking fluid (Model BMF-US, Shelley Medical Imaging Technologies, Ontario, Canada) with a pulsatile carotid artery flow profile and flow rate of 12
The RF signal obtained from the custom designed probe is preprocessed to improve SNR. Transient noise components outside the signal bandwidth are removed by using matched filter followed by a 5th order band pass filter with cut off frequency 1-5 MHz. In addition, a non-linear gain is applied to suppress high amplitude reflections from skin gel interface. Subsequently, the envelope e(t) of the gain compensated signal 𝑥(𝑡) is calculated using Hilbert transform.
1899
e(t) =√𝑥(𝑡)2 + 𝐻{𝑥(𝑡)}2
(1)
dτ where H {𝑥(𝑡)}= ∫∞∞ x(τ) t−τ
RF data of the present frame and that of the next frame to determine the shift in the near wall location.
Further, the envelope is smoothed using a low pass filter with cut off frequency of 10 Hz. C. Wall Detection Typically, acoustic impedance differences produce strong reflections from interfaces. In most conditions, the proximal and distal walls of the artery (phantom tube channel) produce echoes with high amplitude. It is our goal to automatically identify these echoes. We achieve this using adaptive peak detection algorithm.
Normalized Amplitude
On the envelope data, peak detection is performed with an adaptive threshold to detect two peaks ( Pn, Pf)representing the near and far wall. These are the primary peaks and represent the echo produced by the vessel walls (Fig. 3).
Pf n
Pn ,n
Similarly, a window (𝑊𝑓 ) is positioned at the far wall. The peak location in the correlation value provides the shift in the near wall location between the 𝑖 𝑡ℎ and 𝑖 + 1𝑡ℎ frame. Similarly, the shift in far wall location is found. The arterial wall motions of near and far wall (𝑑𝑛𝑤 and 𝑑𝑓𝑤 ) are estimated by cumulatively adding the shifts. Thus the distension waveform is computed as ∆d = 𝑑𝑓𝑤 - 𝑑𝑛𝑤
(2)
The DC drift component was associated with the distension waveform because of the cumulative addition of error in shift computation. Hence, this is removed by applying a 2nd order high pass filter with cut-off frequency of 0.5 Hz. E. Estimation of PWV from Distension Waveform Further, the distension waveforms at multiple adjacent location has been determined from the data acquired by different elements of the transducer, the temporal shift among these waveforms is evaluated to compute the PWV. It was observed that rising edge or wave front of the pulse is the smoothest and most consistent of each pulse. It suffers the least from reflected waves that the pulse wave creates [7]. So, a shift computation by cross correlation of the rising edges of two distension waveforms is performed.
Frame numbers
Figure 3. Primary peaks detected on representative echo
D. Distension Computation Further, on identification of near and far wall the gates are positioned around the echoes, and echo tracking is performed using cross correlation. As an artery expands and contracts in response to the heart beat, the echoes from the artery walls, shifts their position with respect to previous received signal frame. A quasi-periodic pattern is observed due to the walls of the artery moving toward and away from the center.
The typical pressure wave generated by the heart is illustrated in Fig. 4. The rising edge known as the systolic rise period is identified as the segment from the trough to crest. The Dicrotic notch is a slight rise in pressure during diastole due to closing of heart valve. The identification of crest and trough in a distension waveform employing simple threshold based peak detection is inaccurate due to variation in amplitude of the pulses; the Dicrotic notch of a pulse may be higher than the systolic peak of another pulse as illustrated in Fig. 5.
The computation of distension is performed by tracking the change in wall location through successive frames. It is carried out by cross correlating the RF data representing the echoes generated by a wall, from two consecutive frames. Subsequent valley detection is performed around the two primary peaks to find their nearest valleys (𝑉𝑛 , 𝑉𝑓 ). A window (𝑊𝑛 ) of length 2|𝑃𝑛 -𝑉𝑛 | is centered at the peak (𝑃𝑛 ) and cross-correlation is performed between the windowed
1900
Figure 4. Representative pressure pulse in an artery
To overcome this problem, a peak detection algorithm is implemented wherein the number of pulses in the distension waveform is determined from the acquisition period and the pulsatile frequency (determined using Fourier Transform). Further, an adaptive threshold is performed to detect peaks equal to the number of pulses in the waveform. Similarly, valley detection is performed to find the trough of each pulse. The segment from one trough location to the next crest is assigned as a systolic rise period. The temporal shift
from the architecture information of the probe) by the temporal shift (t) computed by the algorithm.
𝑃𝑊𝑉 =
𝐿(𝑒1 ,𝑒2 )
(3)
𝑡(𝑒1 ,𝑒2 )
III. RESULTS AND DISUSSSION The algorithm was implemented on the data acquired by a customized probe from a phantom setup simulating the carotid artery flow profile; the PWV was computed to be 9.6 m/s. Though, the validation using exact ground truth value is difficult as there is no current localized PWV validation technique available. As the proposed approach is based on non-imaging technique it lends to low cost solution for early detection of atherosclerosis. This approach needs to be further validated on the human subjects. IV. CONCLUSION
Figure 5. Distension waveform
between the distension waveforms acquired by two different transducer elements (element 1and element 8) is computed by correlating the systolic rise periods of waveform with their nearest respective systolic rise periods in the next waveform (Fig. 6).
This study proposes preliminary proof of concept for measurement of localized PWV using single 2D ultrasound probe. This provides an opportunity to build additional cardio vascular application on top of the existing ultrasound system without any systemic changes in the design and approach. Our future work will focus on exhaustively validating with data acquired from human subject to analyze it performance and accuracy as the current approach has been carried out on phantom setup to understand the technical feasibility.
REFERENCES [1] Singh, R. B., Sharma, J. P., Rastogi, V., Raghuvanshi, R. S., Moshiri,
[2] [3] [4]
[5] [6] [7] Figure 6. Distension waveform acquired by element 1 and element 8 of the customized probe
Finally, the PWV is calculated by dividing the distance (L) between the two elements under consideration (obtained
1901
M.,Verma, S. P., and Janus, E. D., 1997, “Prevalence of Coronary Artery Disease and Coronary Risk Factors in Rural and Urban Populations of North India,”Eur. Heart J., 18, pp. 1728–1735. Wilkinson IB, Cockcroft JR, Webb DJ (1998). "Pulse wave analysis and arterial astiffness". J. Cardiovasc. Pharmacol. 32 (Suppl 3): S33– 7. PMID 9883745 Hermeling E1, Reesink KD, Reneman RS, Hoeks AP. Measurement of local pulse wave velocity: effects of signal processing on precision. Ultrasound Med Biol. 2007 May;33(5):774-81 Laurent S, Cockcroft J, Van Bortel L, Boutouyrie P,Giannatasio C, Hayoz D, et al 2006 Expert consensus document on arterial stiffness: methodological issues and clinical applications Eur Heart J 2006 22 2588–2605 Bramwell JC, Hill AV (1922). "Velocity transmission of the pulse wave and elasticity of arteries". Lancet 199 (5149): 891–2. Boutouyrie P, Briet M, Collin C, Vermeersch S, Pannier B (February 2009). "Assessment of pulse wave velocity". Artery Research 3 (1): 3–8 Isabelle Vivodtzev et.al., "Arterial Stiffness in COPD", Recent advances in chest medicine, April 2014, Vol. 145(4), pp. 861-875. doi:10.1378/chest.13-1809.
