Local Measurement of the Pulse Wave Velocity using Doppler Ultrasound. Minnan Xu

Transcription

1 Local Measurement of the Pulse Wave Velocity using Doppler Ultrasound by Minnan Xu Submitted to the Department of Electrical Engineering and Computer Science in partial fulfillment of the requirements for the degree of Bachelor of Science in Electrical Engineering and Computer Science and Master of Engineering in Electrical Engineering and Computer Science at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY May 24, 2002 c 2002 Minnan Xu, All rights reserved. The author hereby grants to M.I.T. permission to reproduce and distribute publicly paper and electronic copies of this thesis and to grant others the righttodoso. Author... Department of Electrical Engineering and Computer Science May 24, 2002 Certified by... David Prater VI-A Company Thesis Supervisor Thesis Supervisor Certified by... Roger D. Kamm Professor, Biological Engineering Thesis Supervisor Accepted by Arthur C. Smith Chairman, Department Committee on Graduate Theses

2 2

3 Local Measurement of the Pulse Wave Velocity using Doppler Ultrasound by Minnan Xu Submitted to the Department of Electrical Engineering and Computer Science on May 24, 2002, in partial fulfillment of the requirements for the degree of Bachelor of Science in Electrical Engineering and Computer Science and Master of Engineering in Electrical Engineering and Computer Science Abstract Cardiovascular disease is the leading cause of death in many developed countries. Arteries of people suffering from this disease become stiff and blocked by fatty deposits. In recent years, non-invasive imaging techniques have been playing an increasingly important role in detecting the development of cardiovascular disease. Several methods focus on the measurement of pulse wave velocity, the velocity at which the pressure wave propagates, because it is directly related to arterial stiffness. The objective of this project is to investigate the feasibility of measuring local pulse wave velocity from the blood flow waveforms acquired by Doppler ultrasound. The proposed method includes the following steps: first acquire flow waveforms by Doppler ultrasound at two locations within the same artery, next detect the delay or difference in arrival time of the flow wave at the two arterial locations, and then calculate the PWV by dividing the length of the arterial segment being imaged by the calculated time delay. Although at the conclusion of this study reliable pulse wave velocity detection is not achieved, the study sheds light on many important issues surrounding this potential application. The project explores how sources of variations such as radial postioning of the probe and noise level affect the accuracy of the delay estimate. Thesis Supervisor: David Prater Title: VI-A Company Thesis Supervisor Thesis Supervisor: Roger D. Kamm Title: Professor, Biological Engineering 3

4 4

5 Acknowledgments This thesis project has been a great academic and personal learning experience. Iwould like to thank everyone who helped me through this thesis project. Here is a partial list of all those who helped me learn. Dave Prater, with whom Ihave worked three summers and a term, is the originator of this project idea. He has always inspired me with his energy and innovation. Prof. Roger Kamm, my MIT advisor, who pointed me to so many useful resources. Guohao Dai, who let me borrow his flow system. Andrew Davenport, who solved all my software problems at work. Tony Vallance and the AQ group at Philips, who made me part of a team at Philips. Jim Michner, who helped me with the PVT part of the project. Bill Fry, who helped with modifications made to the Doppler board. Jodie Perry, McKee Poland, and Kim Robertson who helped me with the ever confusing scanner part of the project. David Clark, who help me understand Doppler ultrasound. Dr. Ivan Salgo, my neighbor at Philips, who greeted me every morning and also shared his knowledge of the medical world. Tony Borges, who helped me settle in from the very first day. Tony Brock-Fisher, who gave me some hard to find Doppler ultrasound test fluid. Prof. Denny Freeman, who came to visit me in Andover. Markus Zahn and Lydia Wereminski from the MIT 6A office, who were invaluable in helping through the 6A and M.Eng programs. Raymond Chan and Dr. Robert Lees from the Boston Heart Foundation, who offered great advice on this project and graduate school. The volleyball group at work who made the end of every week a blast. My family, who are supportive always. Iwould also like to thank my family for pushing me to work hard before Iknew any better. And last but not least, Kevin Wilson, my friend in every way. He helped with every part of this project, from discussing ideas to providing moral support and encouragements. Thank you! Minnan 5

6 6

7 Contents 1 Introduction ProblemDescription OverviewofThesis Background ArterialStiffnessandPulseWaveVelocity Existing Methods of Measuring Pulse Wave Velocity FactorsInfluencingPulseWaveVelocity OtherUsesofPWVmeasurement BloodFlowintheCommonCarotidArtery UltrasoundinMedicine DopplerUltrasound Doppler Principle MeasuringBloodFlowwithDopplerUltrasound PulsedDopplervs.ContinuousDoppler PrinciplesofPulsedDoppler Methodology SystemModifications Scanner DopplerProcessor VideoDisplay TimingChangesDuetoSystemModifications

8 3.2 FlowSystemSetup Data Processing DataAcquisition EnvelopeDetection MaximumFrequencyFollower DifficultiesinEnvelopeDetection WindowingtheData DelayEstimation CorrelationBasedTechniques PhaseDifferenceDelayEstimation Results SimulatedData VarianceoftheDelayEstimates RandomNoise FlowSystemData PhysiologicalData Discussion EstimatedPWV SimulatedData FlowSystemData PhysiologicalData HighVarianceintheDelayEstimate Waveform Variation Due to Radial Position Waveform Variation Due to Wave Dispersion Variability Due to Scanning Physiological Variability NumberofHeartCyclesNeeded SystemDelays

9 6.3 VarianceDuetoNoise ResultsofSimulatedNoise NoiseinPhysiologicalData SystemLimitations DualBeam ResolutionofDelayEstimate VelocityMeasurementAccuracy SmallImageBufferSize DelayEstimationMethods FutureWork SystemModifications UserFeedback OtherApplicationsofDualBeamSetup AreasforFurtherResearch Conclusion

10 10

11 List of Figures 2-1 UltrasoundBeamandArtery A Single Spectrum Acquired Using Doppler Ultrasound IllustrationofthePulsedDopplerPrinciple General data path from acoustic signal to video display DiagramofDopplerDetectorBoard Diagram of the Modified Doppler Detector Board Two Spectra Acquired Simultaneously Using Doppler Ultrasound Doppler Signal Path in Non-Duplex Systems FlowSystemSetup Video Display of Dual Beam Phantom Data AcquisitionofDualSpectra ExtractedEnvelopes Low-passFilteringofEnvelopes Windowing TheDelayEstimationProblem AveragePhasedifferences VarianceofPhasedifferences SimulatedEnvelopeData VelocityProfileatCenteroftheArtery Delay Estimates From Different Radial Positions SimulatedDatawithNoiseofSNR=74dB

12 5-5 SimulatedDatawithNoiseofSNR=62dB SimulatedDatawithNoiseofSNR=47dB PhantomFlowWaveEnvelopes FlowSystemDataSet FlowSystemDataSet Physiological Data Set A Physiological Data Set A Physiological Data Set B Physiological Data Set B Physiological Data Set B Physiological Data Set B BeamPlane

13 List of Tables 3.1 ModifiedTriggeredModeFrameTable ModifiedDuplexModeFrameTable DopplerTestFluid SummaryofSimulatedDataSets SimulatedDataParameters EffectofRadialPositioning SummaryofFlowSystemDataSets SummaryofPhysiologicalDataSets TimedelaysandEstimatedPWV

14 14

15 Chapter 1 Introduction Cardiovascular disease is the leading cause of death in many developed countries. Arteries of people suffering from this disease become stiff and blocked by fatty deposits. In recent years, non-invasive imaging techniques have been playing an increasingly important role in detecting the development of cardiovascular disease. Several methods focus on the measurement of pulse wave velocity, the velocity at which the pressure wave propagates, because it is directly related to arterial stiffness. The objective of this project is to investigate the feasibility of measuring local pulse wave velocity from the blood flow waveforms acquired by Doppler ultrasound. Although at the conclusion of this study reliable pulse wave velocity detection is not achieved, the study still sheds light on many important issues surrounding this potential application. 1.1 Problem Description The proposed method is to acquire two blood flow waveforms using a single ultrasound probe at two locations of an artery. The delay in the arrival time of the flow wave at the two points of measurement divided by the distance between the two points gives the pulse wave velocity. The distance of travel is determined by the separation of the two apertures at the two ends of a linear transducer. The details of estimating the delay will be the main focus of the study. 15

16 1.2 Overview of Thesis This thesis is organized as follows: Chapter 2 begins with a brief overview of arterial stiffness, pulse wave velocity, and Doppler ultrasound. Chapter 3 presents the experimental setup and system modifications made for data acquisition. Chapter 4 develops the various steps and algorithms involved in delay estimation. Chapter 5 shows the results from simulated data, flow system data, and physiological data. Chapter 6 analyzes the results and the proposed technique for the detection of pulse wave velocity. 16

17 Chapter 2 Background This chapter presents a brief introduction to arterial stiffness, blood flow, pulse wave velocity, ultrasound imaging, Doppler ultrasound, and the use of Doppler ultrasound to measure blood flow. Section presents current and representative methods of measuring pulse wave velocity and their disadvantages that the proposed method addresses. 2.1 Arterial Stiffness and Pulse Wave Velocity Pulse wave velocity has long been attractive as a diagnostic tool. It is generally agreed that many cardiovascular disorders are associated with increasing rigidity of the arterial wall due to arteriosclerosis 1 [1]. The relationship between the pulse wave velocity (PWV)andthe elasticity of a thin-walled elastic tube filled with an incompressible fluid is expressed by the Moens-Korteweg Equation (2.1). PWV = Eh ρd (2.1) From this equation, we see that the PWV (m/s) is related to the square root of Young s modulus of elasticity (E). Therefore measuring the pulse wave velocity leads to an estimate 1 Arteriosclerosis is a chronic disease characterized by abnormal thickening and hardening of the arterial wall. 17

18 of the stiffness of the tube. Higher velocity corresponds to higher arterial stiffness. The blood density, ρ, of a person should stay fairly constant. The other two parameters h and D may be estimated from the B-mode images of the artery [7] Existing Methods of Measuring Pulse Wave Velocity Pressure Wave or Flow Wave There are existing methods which measure PWV by using Doppler ultrasound [26, 6, 14]. These methods use continuous Doppler and measure two sites very distant in the arterial tree, so these methods have the problem of measuring an averaged value for the PWV. (The problem with acquiring an average PWV value is discussed later.) However these methods demonstrate that the flow wave measured by Doppler ultrasound can be used just as well as the pressure wave for detecting the PWV [1]. Delay Detection Many computer algorithms have been developed to detect the delay between the waves captured at distant sites of the arterial tree [5]. Most of these algorithms rely on detecting the PWV by first identifying characteristic points of the waveforms. The characteristic points are usually chosen near the foot or the lowest point of the flow wave. The foot is preferred because it is relatively free of arterial wave reflection so that error in the calculation of the forward pulse wave velocity is minimized [16]. The technique of finding the time delay between the feet of waveforms works well when the two points of measurement are far apart in the arterial tree making the delay large. This technique would not work for measuring the delay between two waveforms captured a few centimeters apart in the arterial tree. This is because the error associated with locating the foot of the flow wave is around the sampling interval (5 ms) which is on the order of the delay being detected in this application. 18

19 Sequential or Simultaneous Measurements The standard method of measuring PWV is to record a proximal 2 and a distal 3 pressure wave at two different sites on the arterial tree [1]. There are several variations that differ in the choice of artery being measured and whether the measurement at the two sites is sequential or simultaneous. Sequential recording is performed when only one probe is available, therefore simultaneous recording of two sites is not possible. The operator first records the flow or pressure wave from the proximal site and then from the distal site. At both sites, the ECG is also recorded. The time differences between the ECG signal and the flow or pressure waveforms at the two sites, T 1 and T 2,yieldsthetimedelay: T = T 2 T 1. The PWV measurement based on sequential methods is not based on the traveling speed of the same wave generated by the same heartbeat, as is possible in simultaneous techniques. Furthermore, sequential methods do not consider the variability between recordings at the proximal and the distal sites. The standard technique is the simultaneous measurement of the pressure wave propagating from the carotid artery to the femoral artery [6]. However, this method has several problems. The distance the pulse wave travels from the carotid to the femoral is hard to determine. A non-invasive, superficial estimate of the traveled distance is the distance covered by skin between the two transducers. It is known that PWV increases from near the heart to the femoral artery. Thus, the PWV obtained from the popular carotid-femoral technique shows their average value. Local Measurement of the PWV is desired because of the local nature of arteriosclerosis. In the early stage of arteriosclerosis, fibrous spots a few millimeters in diameter are scattered on the arterial wall. In the final stage of arteriosclerosis, after these spots grow, the arterial wall becomes homogeneously hard [6]. Therefore it is important for early diagnosis to measure the local hardness of the arterial wall. It has been shown that health of the certain arteries, such as the carotid artery is highly correlated to atherosclerosis 4 [2]. Therefore it 2 A proximal point is one located toward the center of the body. 3 A distal point is one located far from the center of the body. 4 Atherosclerosis is an arteriosclerosis characterized by the deposition of fatty substance in and fibrosis of the inner layer of the arteries. 19

20 would be beneficial to measure arteries locally. The proposed technique of simultaneously measuring two points in the same artery solves the above problems. The distance of wave propagation is easier to determine since the transducer length and the aperture size are known. The distance between the center of the two apertures is the traveled distance. Measuring two points in the same artery ensures the local measurement of PWV instead of an average value of PWV of two distant locations in the arterial tree. Local PWV provides a measurement of local arterial properties, therefore allowing possibly early diagnosis of arteriosclerosis. Arterial Wall Motion Detection for PWV Measurement As the pressure pulse travels through an arterial segment, the arterial radius at a fixed location expands and contracts from its undisturbed size. Chubachi et al. [6] studies the use of Doppler ultrasound to capture the motion of the arterial wall at two sites. The radio frequency (RF) signal captured from the two sites are used to estimate the difference in arterial wall vibration and hence the propagation delay of the pulse wave. This method requires accurate tracking of the arterial wall from the B-mode ultrasound image, which may be difficult since the arterial walls are very thin Factors Influencing Pulse Wave Velocity Pulse wave velocity can change with many physiological parameters. Age and cardiovascular health are two factors which are important for the basis of this project. As a person ages, his arteries harden gradually and the PWV of an arterial segment increases gradually following a relatively smooth curve. When a person develops cardiovascular problems, the PWV deviates from the normal curve. The proposed technique for PWV measurement has the potential to pick up deviations from the normal curve of PWV increase. The intention is that the technique is simple enough to be used as part of a routine check-up. Each measurement can be seen as a point on the curve of a person s cardiovascular health. If a person s arteries are hardening normally with age, then there is not much reason for alarm. However if a person s arteries are hardening abnormally with disease, having a record of how the PWV is changing 20

21 may help to identify this health problem. Other factors which influence the pulse wave velocity are shorter in time frame, which can be problematic for the measurement of PWV. Studies have shown that the pulse wave velocity varies with respiration [1]. PWV is slightly higher during expiration than inspiration, due to the fact that blood pressure is slightly increased during the expiratory phase. The observed differences of PWVs between the inspiration and expiration phases were less than 0.5 m/s in normal subjects. Typical PWV for the common carotid is from 6.80 m/s to 8.30 m/s. Studies have also shown that after a meal there is a significant increase of PWV in peripheral vessels such as the carotid [11]. The reason for this change has not been established. These factors should be taken into consideration when comparing PWV measurements taken at different times. Blood flow velocity (average of 0.25 m/s in the carotid artery) is small compared to the pulse wave velocity. Therefore correction for the velocity of blood flow itself is small. However any considerable increase in the velocity of the blood as a result of local or general disturbances will cause an equal increase in the velocity of the pulse wave [1]. The study pointed out that any experimentally determined wave velocity must represent the velocity of the wave relative to the blood, plus the velocity of the blood in the artery Other Uses of PWV measurement Aside from using PWV to diagnose the presence of cardiovascular disease, it can also be used to monitor the effect of drugs. Asmar [1] discusses the use of PWV to monitor the effects of antihypertensive drugs used to treat arteriosclerosis, the effect of hormone replacement therapy on arterial properties, and the effect of other possible treatments such as specific food intake and exercise on cardiovascular health. 2.2 Blood Flow in the Common Carotid Artery The common carotid artery was chosen for this study for the following reasons. The carotid artery is easily accessible since it is located close to the skin with no other major arteries 21

22 nearby. The common carotid is straight and can be approximated well by a large straight elastic tube. Health of the carotid artery is important. It has been shown that carotid health is highly correlated to atherosclerosis [2]. The diameter of the common carotid in adults range from 0.2 cm to 0.8 cm with an average value of about 0.7 cm. Typical carotid pulse wave velocity in human ranges from 6.80 m/s to 8.30 m/s [1]. Like other main arteries, the common carotid has a flexible wall that is thin compared to its diameter. The wall expands and contracts in response to pressure pulses. Blood flow can be modeled as pulsatile flow in an elastic tube. Womerley s model of pulsatile flow is presented here. This model is used to generate simulated data (presented in Section 5.1). Womersley s model is for a fully developed pulsatile flow in a straight circular cylinder. This model assumes that the flow is at a location sufficiently distant from the inlet, where the radial and circumferential components of velocity and pressure vanish [22]. The solution, under the above assumption, is shown in Equation 2.2, W (r, t) = 2B [ ( ] 0 r πr R) 2 N B n 1 J 0(α n πr 2 n=1 r R i3/2 ) J 0 (α ni 3/2 ) 1 2J 1(α ni 3 /2) α ni 3/2 (α ni 3/2 ) einωt (2.2) W (r, t) is the velocity profile. R is the radius of the cylinder, J 0 and J 1 are Bessel functions of the first kind of order 0 and 1, respectively. α n = R (nω)/ν where ν is the kinematic viscosity. The non-dimensional parameter α = R ω/ν is known as the Womersley number. Section 5.1 shows plots of waveforms generated by using this model. 2.3 Ultrasound in Medicine In recent years, non-invasive imaging techniques such as ultrasound have been playing an increasingly important role in clinical settings [4]. Ultrasound imaging s moderate cost and capability to acquire data sets over both space and time make it the modality of choice for many diagnostic applications. 22

23 The ultrasound B-mode image is generated by the use of a ultrasonic transducer held up to the body. The transducer transmits ultrasound pulses into the body and receives reflected echoes. As the transmitted signal travels through the body and encounters structural boundaries, part of the signal is reflected. The amount of the ultrasound signal reflected is proportional to the difference in the tissue s acoustic densities. An example of a B-mode image is the small image of the carotid artery at the upper right corner in Figure Doppler Ultrasound Doppler ultrasound is a technique for making non-invasive velocity measurements of blood flow. The transmitter sends out ultrasound pulses. The receiver, instead of measuring how much energy is reflected back by structures in the body, listens for how the signal has changed in frequency Doppler Principle The Doppler principle is utilized by transmitting a signal into the body and observing the changes in frequency that occur when it is reflected or scattered from the targets [7]. When a known frequency is sent out, a moving target returns that frequency shifted by an amount proportional to its velocity. The frequency shift occurs only for the motion that is in the direction of the ultrasound beam. The equation for the Doppler shift is f s = 2f tv cos(θ) c (2.3) where f t is the frequency transmitted, f s is the frequency shift, v is the velocity of the target, c is the velocity of sound in tissue, and θ is the angle between the blood velocity vector and the direction of wave propagation as shown in Figure 2-1. The term cos(θ) shows that only the component of the velocity in the direction of ultrasound propagation is measured. 23

24 v sinθv Direction of Blood Flow v cos θ θ Ultrasonic Beam Figure 2-1: Ultrasound Beam and Artery The ultrasound beam intersects the artery at an angle θ. The component of the blood velocity vector v in the direction of the beam is measured. As the angle increases, the Doppler shift frequency decreases Measuring Blood Flow with Doppler Ultrasound In Figure 2-2, a typical Doppler spectrum of the carotid flow shows the distribution of frequency shifts of velocities of the scatterers in the blood. The spectrum changes with time since blood flow is pulsatile. The shape of the spectrum gives an idea of how blood flow changes temporally. Doppler assessment of blood flow has become routine in many diagnostic ultrasound exams [20]. The use of Doppler today ranges from assessing blood flow in the fetus and umbilical cord, to flow patterns through valves in the heart or monitoring of blood flow to the brain [7]. A variety of instruments are available, ranging from simple pocket-size versions that give an audio output, to very expensive imaging systems that integrate blood velocity measurements with images of anatomy [20]. 2.5 Pulsed Doppler vs. Continuous Doppler There are two modes of operation for Doppler ultrasound, continuous wave (CW) and pulse wave (PW) ultrasound. In CW mode, the transducer has the transmitter and receiver 24

25 Figure 2-2: A Single Spectrum Acquired Using Doppler Ultrasound Spectral Doppler measurement of blood velocity in the center of a common carotid artery. The vertical tick marks in the middle of the spectrum indicates a duration of one second. The vertical axis is proportional to the Doppler shift frequency and in this case has been converted to velocity (units of cm/s). mounted side by side. The transmitter continuously sends out a beam of ultrasound and the receiver is continuously receiving the returned ultrasound signal. In PW mode, a short burst of ultrasound is transmitted at a repetition frequency of f r. The returned signal is received with the same transducer at a time delay of T d after the transmission of the pulse. T d is determined by the range of interest. For PWV measurements, continuous wave (CW) Doppler is traditionally used because of its economic advantage. However, CW Doppler has one major drawback; there is no range resolution. This is because the beam is continually transmitted and received, so the blood motion along the entire beam is being observed. In the PW mode, only the blood motion in the range of interest is observed. In this project, PW is used for its range resolution and also 25

26 because it is easier to make the system changes necessary to acquire two Doppler spectra. 2.6 Principles of Pulsed Doppler Pulsed Doppler provides localized velocity measurements. The instrument transmits a pulse that can vary from a single pulse to 40 cycles long [20], depending on the desired length of the sample volume. The returned signal contains both amplitude and phase information. The phase information can be extracted by coherent demodulation by comparing the received pulses to the reference signal which is oscillating at the frequency sent out. This process is illustrated in Figure 2-3. Short bursts of ultrasound are transmitted at regular intervals, the pulse repetition frequency (PRF), toward a moving target. As the target moves away from the transducer, the returning echoes gradually shift in phase relative to the reference wave. The pulsed Doppler system is sampling this phase shift at the PRF. When the target is moving too fast, or the phase shift occurs too fast, the frequency shift is aliased. Thus, the highest frequency shift pulsed Doppler systems can detect is PRF. 2 The PW sample volume depends on the combination of the transmitted pulse length and the length of the gated range [7]. The width of the sample volume is determined by the width of the beam at the position of the range gate. However, PW sample volumes have not been studied in detail [7]. The distance from the transducer to the beginning of the range cell, Z 1,isgivenby: Z 1 = c(t d t p )/2 (2.4) where c is the velocity of the ultrasound in tissue, t p the pulse length, and t d thetimedelay between the start of transmission and the moment at which the receiver gate opens. The distance from the transducer to the end of the range cell, Z 2,isgivenby: Z 2 = c(t d + t p )/2 (2.5) 26

27 where t p is the period for which the gate is open. The length of the range cell Z r,may therefore be written as the following: Z r = Z 2 Z1 =c(t g + t p )2 (2.6) The sample volume is an important consideration for the maximum frequency envelope and the effect of radial positioning of the probe (Section 6.2.1). Transmit Pulse Receive Pulse Phaser Estimate Blood Motion f s Figure 2-3: Illustration of the Pulsed Doppler Principle Estimates of the blood velocity is derived from a series of phase shifts resulting from motion of blood. f s denotes the Doppler shifted frequency. This diagram is adopted from Hwang [10]. 27

28 28

29 Chapter 3 Methodology This chapter presents the system modifications made to acquire flow waves at two locations in an artery and the flow system set up for simulating pulsatile blood flow. 3.1 System Modifications The underlying system used in this study is the SONOS 5500 from Philips Medical Systems Cardiac Monitoring Group (previously part of Agilent Technologies and which was previously part of Hewlett-Packard.) To make the dual spectrum measurement, the system was modified in several ways. The modifications allow the acquisition of blood flow information from two locations within the artery, then process the incoming Doppler signal as signals from two locations instead of one. Figure 3-1 shows the overall diagram of the data path. Each of the three blocks will be explained in more detail in the following sections. Figure 3-5 shows the data path in a more pictorial format Scanner The scanner environment sets up the the transducer for transmitting and receiving ultrasound pulses. The transducer has 288 piezoelectric elements at its surface (spanning 5 cm). Before every ultrasound beam is shot, only the relevant elements are activated. The correct delay coefficients are also put in place to receive the reflected signal at the appropriate focus. 29

30 Time domain acoustic data Scanner board DDET board PVT card On-screen Display Figure 3-1: General data path from acoustic signal to video display The scanner board, Doppler detector board (DDET) and the PVT card were modified to process the incoming acoustic data captured from two apertures instead of one. For this study, the scanner environment is modified to transmit two pulses, one at each end of the linear transducer. Aperture sizes of 128 elements at each end of the linear probe are used to send out the two pulses. The two pulses are identical (in frequency, direction, and depth) except for their aperture positions. The center of the apertures are separated by a distance of approximately 3 cm. A side effect of alternating between two pulses is that the pulse repetition frequency has been cut by half. Thus the highest frequency shift or velocity detected is decreased by a factor of two Doppler Processor The Doppler processor processes the incoming acoustic signal. The Doppler board consists of three processors: A, B, and C. Processor A wall-filters the acoustic signal to remove low frequencies associated with slow motions of the arterial walls and other nearby tissues. Processor B calculates the spectrum from the acoustic signal and further processes it for video display and audio output. Processor C holds the image and audio data for retrieval. Figure 3-2 shows the data path in the Doppler processor B. The incoming data stream is first windowed by a 128-point Hamming window, then fast Fourier transformed (FFT) to give the the frequency content of the acoustic signal. The spectrum is then FFT shifted so that the the spectrum is ordered by increasing frequency. Next, the spectrum undergoes temporal smoothing. Finally the spectrum is interpolated and gain corrected for video display. The modifications made to the Doppler processor occurs for most of the steps outlined 30

31 DDET Processor B Time domain doppler signal Window FFT Magnitude Detector FFT shift Temporal Smoothing Filter Interpolation to Display Map Gain and Output Figure 3-2: Diagram of Doppler Detector Board Doppler processor B is responsible for processing the incoming acoustic signal into Doppler spectra and prepares the spectra for video display. above. The input data is assumed to be interlaced, with values alternating from the two side of the transducer. The interlaced data is split into two streams before undergoing wallfiltering in Doppler processor A. The result of the wall-filter is interlaced again so that the input to Doppler processor B is of the same format as in the unmodified system. The data stream is again split into two in Doppler processor B before undergoing windowing, FFT, etc. Figure 3-3 shows the modified Doppler processor B. The two streams are put side by side before the interpolation step, with data from one side of the probe on top and data from the other side of the probe on the bottom. This concatenation ensures the correct interpolation to fit the video display area. The FFT size has been changed from the previous value of 128 to 64. Modified DDET Processor B Window FFT Magnitude Detector FFT shift Temporal Smoothing Filter Time domain doppler signal Interpolation to Display Map Gain and Output Window FFT Magnitude Detector FFT shift Temporal Smoothing Filter Figure 3-3: Diagram of the Modified Doppler Detector Board The modified DDET board splits the incoming acoustic signal into two paths and processes them separately producing two Doppler spectral displays. 31

32 3.1.3 Video Display The video display step is responsible for updating the Doppler spectrum on the monitor and drawing the accessory information such as the spectral base line and various markers. The video display is changed to be able to show both spectra (Figure 3-4), each one half the size as the normal single spectrum. Two base lines are drawn instead of one to designate the place of zero frequency shift. More work is needed to make the velocity or frequency shift scales correct for the two spectra. The current velocity scale shows the correct velocity range for one of the two spectra. The B-mode image in the upper right shows one gate for the beam corresponding to one end of the probe. Due to time constraints, the software changes necessary for displaying the other gate was not implemented. Figure 3-4: Two spectra acquired simultaneously using Doppler ultrasound The small window in the upper right shows the B-mode image of the carotid artery. The circle designates one of the sample volumes being imaged. The other sample volume, which is not drawn, lies at the other end of the artery parallel to the first sample volume. The bottom display shows the two spectra. The upper spectrum corresponds to the flow waveform captured at the marked sample volume. The lower spectrum corresponds to the flow waveform captured at the unmarked sample volume. 32

33 3.1.4 Timing Changes Due to System Modifications There are two modes of operation in the PW system: non-duplex and duplex. The nonduplex or Triggered Mode continuously updates the Doppler waveform. The B-mode image at the upper right of the video display is not updated until the user requests B-mode imaging. Thus at a given time only one of the two displays (Doppler or B-mode) is active. When the Doppler display is active, the part of the frame table (shown in Table 3.1) is repeated until an interrupt occurs. The interrupt brings the system to another part of the frame table where only 2D lines are shot. The Duplex Mode keeps both the Doppler display and the B-mode image active. The frame table consists of repeats of the fragment shown in Table 3.2. The frame table of both the Triggered Mode and Duplex Mode are modified to accommodate two Doppler lines shot from different aperture settings. Tables 3.1 and 3.2 shows the altered versions. The Doppler line time of µs is only one possible value. The Doppler line time depends on the gate depth and other system parameters. Figure 3-5 shows how how the non-duplex frame table is repeated to give the Doppler acoustic signal. The Doppler signal is continuously windowed. Each window results in one spectral line. Since the windows are overlapping, each FFT carries redundant information. This redundancy ensures that the Doppler spectrum appears relatively smooth in time. The number of new acoustic samples captured by each window changes with the PRF, allowing the window length, FFT length, and display size to remain constant. 33

34 Line Type Line Time (µs) Doppler Setup (S) Doppler (A) Doppler (B) Table 3.1: Modified Triggered Mode Frame Table The modified Triggered Mode frame table allows two pulsed Doppler beams to be shot alternating from two ends of the transducer. The Doppler setup line loads the correct delay coefficients to receive the reflected signal at the appropriate focus. The Doppler line times depend on the PRF. The values given are only a possible value. Line Type Line Time (µs) 2D Load Line (2D) Doppler Setup (S A ) Doppler (A) Dop Setup (S B ) Doppler B (B) Table 3.2: Modified Duplex Mode Frame Table The modified Duplex Mode frame table allows two pulsed Doppler beams to be shot alternating from two ends of the transducer. The 2D load line is one of the lines used to build the B-mode image. The 2D line has an associated angle which changes with each repetition of the frame table. The Doppler setup lines loads the correct delay coefficients to receive the reflected signal at the appropriate focus. The Doppler line times depend on the PRF. The values given are only a possible value. 34

35 33.61µs 1/PRF system delays S A B S A B S A B Repeat of Frame Table (Scanner Environment) A A A B B B Acoustic signal in time (Doppler Detector Environment) FFT Doppler Spectra Display Area Spectral Display (PVT Environment.) 5 ms Figure 3-5: Doppler Signal Path in Non-duplex systems The scanner follows the frame table shown in Table 3.1. S is the setup time during which the scanner loads the correct delay coefficients to receive the reflected signal at the appropriate focus. A and B are Doppler lines from the two ends of the transducer. The time between 2 Doppler lines from the same aperture is 1/PRF. The system delay is the time between 2 Doppler lines from different apertures. There are two possible system delay times: one after A is shot and before B is shot, the other after B is shot and before A is shot. Both the PRF and the system delay depend on the time needed to shoot one Doppler line. This time is determined by the depth of the gate and other system parameters. 35

36 3.2 Flow System Setup A flow system or flow phantom was used to measure the pulse wave velocity in a more constrained environment. An elastic tube was hooked up to a bicycle pump. An illustration of the setup is shown in Figure 3-6. The flow system was used to test the feasibility of detecting delays, and how velocity profiles change with radial positions. The flow system consists of two fluid reservoirs, one higher than the other. When pressure is applied, fluid flows from the lower reservoir through the elastic tube and then to the higher reservoir. When pressure is released, fluid from the higher reservoir flows back to the lower reservoir without flowing through the elastic tube. A hard plastic housing holds the tube at two ends. When imaging, the entire housing and the elastic tube are submerged in water. The ultrasound probe is fixed about a half centimeter above the elastic tube with water in between the probe and the tube. The elastic tube is roughly 14 cm in length and 0.5 cm in diameter. The fluid used is a blood-mimicking fluid from ATS Laboratories. Table 3.3 shows the fluid specifications. A bicycle pump provides the pressure required to drive the fluid through the flow system. Resistance to the flow system is accomplished by a screw mechanism. At maximum resistance, the tube can be completely blocked of flow. Density 1.04 g/cc Viscosity 1.66 centistokes Particle Size 30 µm Particle Concentration 1.7 per cc Table 3.3: Composition of the Doppler test fluid used in the flow system. An example of the signal captured from the flow system is shown in Figure 3-7. The signal quality depends much on the scatterers in the test fluid. To achieve better signal quality, it is important to shake up the test fluid before using to make sure enough particles are in the fluid to scatter the ultrasound pulses. 36

37 Reservoirs One-Way Connector Apply flow resistance here Probe Tube Elastic Tube Tube Housing Resistance Drawing not to scale Pump Figure 3-6: FlowSystem Setup A flow system was used to measure the pulse wave velocity in a more constrained environment.the flow system consists of two fluid reservoirs, one higher than the other. When pressure is applied, fluid flows from the lower reservoir through the elastic tube and then to the higher reservoir. When pressure is released, fluid from the higher reservoir flows back to the lower reservoir without flowing through the elastic tube. A hard plastic housing holds the tube at two ends. When imaging, the entire housing and elastic tube is submerged in water. The ultrasound probe is fixed about a half centimeter above the elastic tube with water in between the probe and the tube. The elastic tube is roughly 14 cm long and 0.5 cm in diameter. The fluid used is a blood-mimicking fluid from ATS Laboratories. Resistance to the flow system is accomplished by a screw mechanism. 37

38 Figure 3-7: Video Display of Dual Beam Phantom Data The data is acquired by imaging the flow system illustrated in Figure 3-6. The quality of the image is highly dependent on the amount of scatterers suspended in the Doppler test fluid. To achieve better signal quality, it is important to shake up the test fluid before using to make sure enough particles are in the fluid to scatter the ultrasound pulses. 38

39 Chapter 4 Data Processing Once the data has been brought off-line, it undergoes many data processing steps before producing a PWV estimate. First the maximum frequency envelope is extracted from the Doppler spectra. Then the more representative part of the envelope is extracted by windowing. Next the delay between the upstream and downstream envelopes are calculated. This delay is then converted to a PWV estimate. For each step, there are many methods available. This section will present the method chosen for each step and why they are more suitable than other methods for this application. 4.1 Data Acquisition Data is acquired with the modified Doppler ultrasound system. Section 3.1 describes the modifications. The acquired data has two flow spectra as seen in Figure 3-4. For each subject, over 40 cardiac cycles are acquired. Image data such as shown in Figure 3-4 is captured and then brought off-line for analysis. Image data is the easiest to acquire. Ideally one should try to access the raw acoustic data. The linear probe is placed on the neck over the carotid artery. The center of the two beams are placed in the middle of the artery, or as close to the center as possible. This is because flow velocities vary with radial position (Figure 4-1). The human subject should be at rest for about 5 minutes prior to measurement to establish steady flow rate [26]. 39

40 The SONOS 5500 has enough buffer size to store roughly 8 cardiac cycles. Each time the buffer is filled up, data acquisition is stopped to store the content of the buffer to disk. After the image data is stored, measurement is resumed. Therefore the data acquired is not from consecutive cardiac cycles, but from sets of 8 cardiac cycles. Probe Figure 4-1: The linear probe is held up to the neck close to where the carotid artery is located. The angle of measurement is away from the head in order to avoid the carotid bifurcation. The focus of the two beams are placed as close to the center of the artery as possible. the arrow shows the direction of blood flow. 4.2 Envelope Detection The Doppler ultrasound signal from blood flow contains a spectrum of frequencies whose amplitude is related to the velocities of blood within the sample volume. Estimating the delay directly from a 2D image is computationally intensive. A one-dimensional signal is more suitable for delay estimation. There are many ways to extract a meaningful 1D signal to capture how the signal is changing with time. Commonly, the highest frequency present in the Doppler signal at a particular time is used to represent the Doppler signal. Studies 40

41 have shown that the maximum frequency envelope is relatively insensitive to changes in beam-vessel geometry [8]. Other possibilities include the mean frequency envelope and the median frequency envelope Maximum Frequency Follower There are various methods of extracting the maximum frequency envelope. In theory, the maximum frequency simply corresponds to a maximum velocity present in the volume of blood flow being imaged. This seemingly simple task is complicated by intrinsic spectral broadening and noise [7]. A number of approaches have been developed to deal with these difficulties. For this project, the threshold-crossing method adaptive to background noise is used [8]. A threshold is determined based on the estimated noise level. The magnitude of each bin of the spectrum (scanning from high to low frequency bins) is compared with the threshold. When in a sequence of r successive bins there are at least m bins which exceed the threshold, then the highest bin frequency in that sequence is assigned as the maximum frequency. Figure 4-2 shows the extracted envelopes from one cardiac cycle. The two spectra correspond to the upstream and downstream locations in the artery Difficulties in Envelope Detection The extracted maximum frequency envelope is very noisy. This is due to several factors. The Fourier transform-based spectrum analyzer exhibits a characteristic granular pattern called Doppler speckle which causes large random fluctuations of the instantaneous spectral amplitude from the true amplitude. The size of the deviation from the true amplitude is comparable to the true amplitude [8]. Other sources of fluctuations are electronic noise and interference from anatomical structures near the artery. The envelope goes through low-pass filtering to remove high frequencies which are considered to be mostly noise. This assumption is explained later in Section Figure 4-3 shows the envelopes of Figure 4-2 after filtering. Filtering introduced a delay corresponding to the filter length. The delay caused by the filter will not affect the accuracy of the propagation delay estimation, since the filter introduces equal amount of delay to both the upstream and the downstream signal. 41

42 80 upstream downstream Velocity (cm/s) Time (Sec) Figure 4-2: Extracted Envelopes Maximum frequency followers are used to extract envelopes from Doppler spectra Velocity (cm/s) Low pass filtered upstream Low pass filtered downstream Time (Sec) Figure 4-3: Low-pass filtering of Envelopse Filtering introduced a delay size corresponding to the length of the low-pass filter. 42

43 4.3 Windowing the Data The rising edge or wave front of the pulse wave is the smoothest and most consistent of each pulse. This part of the wave suffers the least from reflected waves that the pulse wave creates [16]. The maximum frequency envelope is smoothest and most consistent at the rising edge of each pulse [7]. In this implementation, only the segment of data surrounding the peak of each cardiac cycle is extracted for delay estimation. Because of the finite length of the extracted segment, it is important to apply a window. Application of the window eliminates the problems caused by rapid changes in the signal at the edges of the window. These problems include misrepresentation of the high frequency information. The window used is a Blackman window with a stretch of ones filled in at the top. Figure 4-4 shows the effect of windowing on the signal. The wave front and the part of the cardiac cycle with high flow has been extracted for analysis upstream window Windowed Upstream Velocity (cm/s) Velocity (cm/s) Time (s) Time (s) downstream window Windowed Downstream Velocity (cm/s) Velocity (cm/s) Time (s) Time (s) Figure 4-4: Windowing The two plots on the left show the upstream and down stream flow waveforms with the window overlaid. The two plots on the right show the result of windowing. The wave front and the part of the cardiac cycle with high flow has been extracted for analysis. 43

44 4.4 Delay Estimation The objective of time delay estimation is to determine the delay between two scaled versions of the same signal, s(t), in the presence of noise, n(t) [15]. This involves sampling and processing two continuous time signals x 1 (t) andx 2 (t) givenby x 1 (t) =C 1 s(t)+n 1 (t) (4.1) x 2 (t) =C 2 s(t D)+n 2 (t) (4.2) In a discrete time sampled data system, with sampling period T, the estimation problem reduces to determining an estimate ˆD of the true time delay D using a finite set of measured samples, x 1 (kt) andx 2 (kt), of the signals x 1 (t) andx 2 (t). Delay estimation for this application is performed on the maximum frequency envelopes of two Doppler spectra of the same cardiac cycle (Figure 4-5). The signals x 1 (kt) and x 2 (kt) are sampled from the continuous time upstream and downstream maximum frequency envelopes respectively. The delay estimate ˆD canthenbemappedtoanestimateofpwv by the following relation: PWV = L ˆD (4.3) where L is the distance of travel or the distance between the two apertures of the probe. The results of Chapter 5 present delay estimates in terms of samples. Each sample corresponds to a delay of T ms. Thus time delay is related to sample delay by the following: ˆD = sample delay T (4.4) 44