Sonoelastography using Compensated Power Doppler
Sonoelastography using Compensated Power Doppler
Stephen J.McKenna,Stuart Dickson,Ian W.Ricketts Department of Applied Computing
University of Dundee
Dundee,Scotland
{stephen,sdickson,ricketts}@https://www.wendangku.net/doc/0317923035.html,
Asif Iqbal,Tim Frank,Alfred Cuschieri Department of Surgery and Molecular Oncology
University of Dundee
Dundee,Scotland
{a.y.iqbal,t.g.frank,a.cuschieri}@https://www.wendangku.net/doc/0317923035.html,
ABSTRACT
Sonoelastography is the visualisation of elastic properties using ultrasound.It can enable tumours to be detected and localised based on their elasticity when they are less elastic than the surrounding soft tissue.In vibration sonoelastog-raphy the target tissues are vibrated while simultaneously recording ultrasound images.A technique for imaging rel-ative elastic properties is proposed that uses a standard ul-trasound machine.It combines B-scan and power Doppler signals to produce images of relative vibration amplitude. Preliminary results using simulations and liver phantoms are presented and the potential of the method to highlight areas of differing elasticity within an organ such as the breast is mentioned.The possibility of combining such a method with freehand3D scanning,enabling B-scan and power Doppler signals to simultaneously populate a voxel array for subsequent visualisation is discussed.
KEY WORDS
Sonoelastography,ultrasound,freehand scanning,power Doppler ultrasound
1Introduction
The most common clinical method for detecting lumps within tissue,palpation,is highly subjective and depen-dent on the skill of the practitioner.The method exists because certain pathological conditions,such as malignant tumours,manifest themselves as changes in the tissue’s mechanical stiffness.While X-ray imaging is well estab-lished for the detection of small,deeply located tumours, X-ray hazards and the desire for better performance have led to a continuing search for alternative techniques.Di-agnostic ultrasound is a potential alternative to X-rays,its limitation being that small pathological changes in tissue are dif?cult to discern on normal ultrasound B-scans.If, however,ultrasonic echo data is collected before and af-ter a slight compression of the tissue,comparisons can be made between normal and pathological areas.This is pos-sible when normal tissues exhibit relatively more move-ment than stiffer pathological regions.It has been suggestd that benign and malignant tumours can be distinguished by elastography(imaging of elasticity)due to their differing uniformity of elastic properties[14].Elastography exam-ines the static elastic properties of tissue.Other authors and unpublished results from our own tissue property stud-ies suggest that differences between healthy and pathologi-cal tissue are highlighted more clearly using rapidly chang-ing strain.This time-dependent(i.e.viscous)response is analogous to the vibrational frequency response of the tis-sue.It is likely that a relatively narrow band of vibration frequencies exists for which the response in the tissue is optimum for distinguishing variations in viscoelastic prop-erties.Ultrasound imaging of elastic properties in the pres-ence of vibration is known as sonoelastography.A small, stiff zone will appear as a de?ned region due to the dif-ference between its motion and that of the surrounding tis-sue.Ultrasound sonoelastography imaging has been com-pared to conventional ultrasound imaging for the detection of prostate cancer in vitro,with promising results[18].Al-though elastography and sonoelastography are not yet be-ing used in routine clinical practice,these imaging methods have the potential to give comparable spatial resolution to standard grey-scale imaging with enhanced tissue discrim-ination.
Several authors have reported the use of phase-contrast magnetic resonance imaging(MRI)to visualise the mechanical properties of tissues[2,5].Images of tis-sue subjected to static or time varying displacement are obtained,yielding information on the3D distribution of elasticity and viscoelasticity respectively.The results from these techniques are very impressive in that small inho-mogeneities can be localised.However,the method may never become broadly applicable due to the very high cost of MRI and its lack of portability.
In Doppler ultrasonography,shifts in returning ultra-sonic echoes due to the motion of re?ecting features are detected.While this is commonly used to highlight blood ?ow,variations in tissue motion can also be revealed.Ul-trasound imaging is particularly attractive as the basic im-age forming process due to its benign nature and low cost. In addition,the colour Doppler imaging modality offered in modern systems provides motion information when vi-bration is present.The technique described here,known as Doppler sonoelastography,produces images that can be interpreted in terms of elasticity variations.
Sonoelastography has the potential to enhance the ef-?cacy of screening for common cancers.A primary moti-vation for conducting the research described here is to lay the foundations for developing a new non-invasive,atrau-
matic and painless system for breast tumour diagnosis and location.The technique is applicable to other common tu-mour sites such as the prostate.Elasticity imaging may also be used to detect other diseases that cause changes in tissue elasticity(e.g.atheromatous disease).Other applications include the measurement of tissue elasticity for use in tis-sue modelling studies and to provide data in virtual reality research in relation to surgical/interventional simulators.
In this paper,we investigate the possibility of using the power Doppler imaging mode available as standard on many modern ultrasound machines in combination with co-registered B-scans to perform sonoelastography.Sec-tion2brie?y describes the methods that have previously been proposed for imaging elasticity using ultrasound.In Section3,a process for estimating the power Doppler sig-nal from tissue under vibration is described and simulated. Simulation results are presented to illustrate the effect of scatterer strength,vibration frequency and vibration ampli-tude on the power Doppler signal.Section4describes pre-liminary imaging experiments using the theory presented in the previous section.Section5describes our freehand3D scanning system and discusses its applicability to Doppler sonoelastography.Finally,some conclusions and directions for future work are given in Section6.
2Elasticity Imaging using Ultrasound Several techniques for imaging tissue elasticity using ultra-sound have been proposed[1,4,10,12,13,20].The reader is refered to Gao et al.[7]for a review.Taylor et al.[19] classify existing methods as(i)compression elastography (strain imaging),(ii)transient elastography and(iii)vibra-tion sonoelastography.In compression elastography,ultra-sound images are compared before and after a compression is applied to the tissue in order to compute a strain map.In transient elastography,a low-frequency transient vibration is applied and the resulting tissue displacement is detected using ultrasound before echoes from tissue boundaries oc-cur.The third class,vibration sonoelastography,which is the topic of this paper,images the vibration patterns result-ing when low frequency vibration is applied to the tissue. Vibration propagation within a complex organ cannot be solved analytically.However,small,stiff lesions will tend to result in decreases in vibration amplitude.The extent to which they are contrasted with the surrounding tissue will depend on their size and stiffness and on the frequency of vibration.Losses at high frequencies impose an upper limit on the vibration frequencies that can be used in practice.
Vibration amplitude imaging with low-frequency vi-bration was?rst proposed in the late80’s[11].Tissue mo-tion can be estimated by tracking the two-dimensional im-age motion of the speckle produced by back-scattering in high frame-rate,real-time ultrasound.Such tracking is of-ten based on template matching methods(e.g.correlation-based motion estimation)although other optical?ow algo-rithms may also be applicable[3].Speckle tracking can be used to produce images of strain magnitude(e.g.[9]).Finite element methods have been proposed to enable re-construction of the spatial distribution of Young’s modulus [4].
An alternative to speckle tracking is the use of real-time Doppler ultrasound.Doppler techniques measure the component of motion in the direction of ultrasound wave propagation and as such detect axial motion.Doppler ultra-sound machines typically use autocorrelation estimators to estimate the mean frequency and the variance of the power spectrum[6].In?ow Doppler,which is used for imag-ing blood?ow for example,the mean frequency is used to estimate the mean velocity.However,under sinusoidal vibration,mean velocity gives no indication of vibration amplitude since oscillation is about a rest position.Taylor et al.[19]make use of the relationship between vibration amplitude and?ow Doppler variance.They used a scan-ner specially modi?ed to display the real-time estimate of the variance of the power spectrum.Under reasonable as-sumptions the standard deviation of the power spectrum is linearly related to vibration amplitude.They were thus able to image the vibration amplitude by measuring this vari-ance.
In contrast,we propose the use of power Doppler imaging in conjunction with co-registered B-scan images to image the vibration amplitude.Power Doppler is available as standard on many ultrasound machines since it is use-ful for imaging blood?ow and has the advantage of good sensitivity to very low velocities.
3Power Doppler
Power Doppler imaging encodes an estimate of the inte-grated power Doppler spectrum in pseudo-colour.It cannot be used directly to estimate the vibration amplitude because the power Doppler signal depends on the echo strength of the region being imaged as well as its vibration amplitude. The method proposed here uses standard B-scan imaging to compensate the power Doppler signal in order to image the relative vibration amplitudes in the tissue.
3.1Simulation
Ultrasound scanners tend to use autocorrelation estimators for colour?ow and power Doppler imaging.This estima-tion process was simulated in order to investigate its be-haviour at different vibration frequencies and vibration am-plitudes.The simulation was similar to that used by Taylor et al.[19]to investigate the effect of vibration amplitude on estimates of the variance of the Doppler power spectrum as used for?ow imaging.Here we investigate estimation of vibration amplitude from the integrated power spectrum.
In pulsed Doppler ultrasound,a sequence of ultra-sound pulses which together form a packet are used.The number of pulses in a packet is usually user-controlled and here it was set to N=16.These pulses are emitted at in-tervals of T p.Each pulse was modelled as a real wavelet
(Equation(1))whereσ=173ns and f c was the centre frequency.
p(t)=
1
(2πσ2)12
exp(?
t2
2σ2
)cos(2πf c t)(1)
The round-trip time taken from the ultrasound transducer
to a scatterer at distance d0and back is t d
0=2d0/c where
c is the spee
d of sound.Th
e mean speed o
f sound in tis-sue varies from approximately1446ms?1in fat to approx-imately1556ms?1in spleen,for example.The simulations reported here used c=1540ms?1,an average value used in some scanners[8].A scatterer under forced vibration was modelled as undergoin
g sinusoidal motion about a rest position at distance d0from the ultrasound transducer wit
h peak vibration amplitude m and vibration frequency f v. Its distance,d(t),from the transducer at time t is therefore given by Equation(2).
d(t)=d0? m sin2πf v t(2) The backscatter,e(t),received from a single pulse at time t is:
e(t)=Ap(t?2d(t)/c)(3) where A is the amplitude of the backscatter and models the echogenicity,the scattering coef?cient and attenuation in the tissue.A?rst-difference FIR?lter was applied to the backscattered pulse signals.This simulated a wall?lter and attenuated the https://www.wendangku.net/doc/0317923035.html,ponent.
Quadrature demodulation was simulated by sampling the N returning,?ltered,backscattered pulses in a packet with the?rst pulse emitted at time t=0.Samples I n
were taken at times t I(n)=nT p+t d
0and samples Q n
were taken a quarter of a cycle later at times t Q(n)= t I(n)+1/(4f c),where n=1...N.In the experiments described here,T p=1ms.The average power,R,of the backscattered signal is given by the autocorrelation func-tion at zero lag:
R=1
N
(I2n+Q2n)(4)
Note that R is proportional to A2.Doubling the amplitude of the backscatter,for example,will quadruple the power Doppler signal.
3.2Simulation Results
Figure1shows power,R,plotted against vibration ampli-tude, m for a scatterer vibrated at a frequency of f v= 200Hz at mean distance d0=2cm from a7.5MHz trans-ducer.Three curves are plotted corresponding to scatterer strengths of A=0.1,A=0.2and A=0.3.Together these curves illustrate that R increases with A2.If the power signal can be compensated using an estimate of the scatterer strength these three curves become the same.The power Doppler signal increases monotonically with vibra-tion amplitude over this range(≤1mm)except at very
Figure1.Simulated power Doppler versus vibration am-plitude for A=0.1(solid),A=0.2(dashed)and A=0.3 (dotted).The power Doppler signal has been normalised. Other parameters were d0=2cm,f c=7.5MHz and f v=200Hz
low power.Therefore an appropriately compensated power Doppler signal could be used to image vibration ampli-tude in this situation.Note that larger vibration amplitudes would not be properly imaged.
The amplitude and frequency of the vibration source should be selected to be appropriate for the ultrasound transducer used and the depth d0of the tissue of interest. This is necessary in order to ensure that the relationship be-tween R and m is approximately monotonic and produces a meaningful image for the range of vibration amplitudes induced.
Figures2and3show three curves generated as in Fig-ure1but with the vibration frequency decreased to40Hz. Note that the ranges of vibration amplitudes that can now be imaged reliably are different.Figure2shows that the range m=0to m=5mm could in theory be imaged reasonably well.However,Figure3shows that vibration amplitudes in the range m=0to m=250μm could also be imaged at this depth and frequency.
4Sonoelastographic Imaging
An Aloka SSD220ultrasound system with a7.5MHz linear array probe was used for preliminary imaging experiments. This system has a power Doppler mode as standard.
Several vibration systems have been constructed for experimental use with tissue phantoms.They all consist of a single source using a signal generator and an acoustic speaker.The oscillator incorporated a trigger circuit to al-low phase-locking of the image acquisition.The range of vibration frequencies that can be used is limited by loss at higher frequencies.Frequencies need to be selected to en-able imaging of the vibration amplitude as described in the previous section.An alternative to sinusoidal vibration is
Figure2.Simulated(nomalised)power Doppler versus vi-
bration amplitude for A=0.1(solid),A=0.2(dashed) and A=0.3(dotted).Parameters were as for Figure1 with the exception of vibration frequency which was f v= 40
Hz.
Figure3.The previous Figure with the abscissa scaled. to use some form of polychromatic vibration(e.g.square wave vibration).This can help avoid modal patterns and the vibration systems are currently being used to conduct empirical investigation of various forms of vibration over a
range of frequencies and amplitudes.
Figure4shows co-registered B-scan and power Doppler images of a slice through a commercially available synthetic breast phantom(supplied by Computing Imaging Reference Systems,Inc.).Vibration was applied at the base in the Figures and the transducer was at the top.The fre-quency of the applied vibration was30Hz.The images were obtained by keeping the ultrasound probe clamped in a?xed position while switching from B-scan mode to power Doppler mode.A?uid?lled cyst can be seen at the lower right and a stiff lesion at the lower left.The cyst ap-pears as a void in the B-scan
and
therefore also as a void
in the power Doppler image.Since there is no signi?cant
backscattered signal from this region,there is no power
Doppler signal.The lesion at the lower left appears con-
trasted in the B-scan indicating that it scatters with greater
amplitude.It also appears contrasted in the power Doppler
signal and the extent of this contrast is due to both the in-
creased scatter and the increased stiffness.
Figure4.Mean B-scan(left)and power Doppler(right)
images of the same slice of a commercial breast phantom
with an applied vibration.
Figure5shows co-registered B-scan,power Doppler
and sonoelastographic images.The sonoelastographic im-
age was obtained by compensating the power Doppler data
with squared B-scan data.The square root of the resulting
pixel data is shown for better visualisation.The images are
of a liver phantom with a simulated tumour in the centre.
This tumour appears hyperechoic in the B-scan but is not
apparent in the power Doppler signal as a result,despite
its lower elasticity.However,it appears as a void in so-
noelastographic image indicating that it is stiffer than the
surrounding tissue.
5Freehand3D Sonoelastography
Standard two-dimensional ultrasound scanners can be used
to provide three-dimensional images if the3D position and
orientation of the transducer is known for each of the2D
images recorded.Free-hand3D scanning can be used to
construct a3D volume(voxel array)while the physician
performs an examination in a normal manner.Visualisa-
tion software can then provide an operator with the ability
to explore the3D volume using techniques such as any-
slice imaging,rendering with transparency and3D view-
point control.
5.1Calibration and Visualisation
A freehand scanning system has been constructed in which
the ultrasound probe’s3D position and orientation are mea-
sured using a Polhemus Fastrak electromagnetic sensor
with an angular and positional accuracy of0.5?and0.5mm
respectively.The Polhemus device is portable and accurate
Figure5.Top:B-Scan,Middle:Power Doppler,Bot-tom:Sonoelastographic image.The sonoelastographic im-age was obtained by compensating the Dopper image with squared B-scan data and then taking the square root of each pixel to improve visualisation.
as long as reasonable precautions are taken to avoid elec-tromagnetic interference.Once attached to the ultrasound probe,the Polhemus sensing system is spatially calibrated with respect to the B-scans using an object of known ge-ometry.This spatial calibration was performed using the StradX software and a phantom constructed using a sim-ilar design to that described by Prager et al.[15].This phantom is shown in Figure6.The measurements from the Polhemus sensor and the ultrasound images captured
were Figure6.Left:the calibration phantom with the probe and Polhemus sensor attached.Right:a calibration image temporally aligned using a trigger signal supplied via a footswitch.This system can be used to reconstruct3D vol-umes from freehand scans and is being used to investigate the possibility of3D sonoelastography based on simultane-ously acquired B-scan and power Doppler data.High qual-ity3D reconstruction requires accurate calibration of the free-hand scanning device,accurate registration and data fusion.Image-based registration and fusion can improve the quality of the reconstruction and reduce speckle noise, shadowing and signal dropout[16,17].
5.2Current Research Issues for3D Doppler
Sonoelastography
Doppler imaging only measures the component of velocity in the direction of wave propagation i.e.only axial motion can be detected.However,freehand scanning can be used to register and fuse Doppler signals from different trans-ducer orientations.
The process of forming a power Doppler image is rel-atively slow since multi-pulse packets are used for estima-tion.This means that the probe can move signi?cantly dur-ing acquisition of a single image.Probe position and orien-tation measurements therefore need to be interpolated over time so that columns in the2D images obtained from the linear array probe can be aligned appropriately in3D.The frame-rate for power Doppler imaging is often increased by reducing the area over which Doppler signals are estimated, the remaining area being imaged in B-scan mode.During such a scan,the B-scan and Doppler data can be recorded and registered in two separate3D voxel arrays.Fusion of these B-scan and Doppler volumes then compensates the power Doppler data to produce3D sonoelastographic vol-ume data.Each voxel in this volume could be assigned an uncertainty value based on the amount of evidence avail-able from the2D scans and the extent of the interpolation used.This uncertainty data could in turn provide useful in-formation for subsequent data fusion and provide feedback to the physician performing the scan.Three-dimensional sonoelastographic imaging has the potential application of measurement of the volume distribution of tissue elastic properties.This data is required by researchers developing mathematical models of tissue and,in particular,for use in electronic tissue representation in surgical simulators.
6Conclusions
A method for vibration sonoelastography has been pro-posed based on compensating power Doppler images with B-scan image data.Simulations were presented to explore the feasibility of the method and the sensitivity to vibration amplitude and frequency.An experimental imaging system was described and the potential bene?ts of its extension to freehand3D scanning systems were outlined.
Future work could extend the simulation using?nite element shear wave vibration modelling combined with models of ultrasound propagation.As demonstrated in the simulations presented here,sonoelastographic scanning re-quires the determination of optimal frequency ranges for the externally applied vibration.Modal patterns(stand-ing waves)can occur,especially for regular shapes,and these can make image interpretation dif?cult[19].The use of polychromatic vibration should be explored to alleviate such artefacts.We are also studying Doppler imaging in the non-power mode.Here the image acquisition times are shorter and allow us to synchronise the image to an ad-justable phase position within one vibration cycle.This is revealing information about vibration mode structures and will help to develop methods for avoiding them.
An objective of this research is to improve detec-tion of subclinical breast cancer by breast sonoelastogra-phy.The sonoelastographic imaging modality is painless, risk free and uses equipment that is relatively cheap and portable.
References
[1]J.C.Bamber and N.L.Bush.Freehand Elasticity Imag-
ing Using Speckle Decorrelation Rate.Plenum,New York, 1995.
[2]J.Bishop,M.Leitch,G.Poole,and D.Plewes.Magnetic
resonance imaging of shear wave propagation in excised tissue.Journal of Magnetic Resonance Imaging,8:1257–1265,1998.
[3] D.H.Cooper and J.Graham.Estimating motion in noisy,
textured images:Optical?ow in medical ultrasound.In British Machine Vision Conference(BMVC),pages585–594,1996.
[4]M.M.Doyley,P.M.Meaney,and J.C.Bamber.Evalu-
ation of an iterative reconstruction method for quantitative elastography.Phys.Med.Biol.,45:1521–1540,2000. [5]R.Ehman.Magnetic resonance elastography:‘palpation’
by imaging.Proc.Int.Workshop on Soft Tissue Deforma-tion and Tissue Palpation,October1998.
[6] D.H.Evans and W.N.McDicken.Doppler Ultrasound:
Physics,Instrumentation and Signal Processing.Wiley,2nd edition,1999.
[7]L.Gao,S.K.Alam,R.M.Lerner,and S.F.Levinson.Imag-
ing of the elastic properties of tissue-a review.Ultrasound Med.Biol.,22:959–77,1996.
[8]J.A.Jensen.Ultrasound imaging and its modeling.In Imag-
ing of Complex Media with Acoustic and Seismic Waves, Topics in Applied Physics.Springer Verlag,2000.
[9]K.Kaluzynski,X.Chen,Y.Emelianov,A.R.Skovoroda,
and M.O’Donnell.Strain rate imaging using two-dimensional speckle tracking.IEEE Transcations on Ultra-sonics,Ferroelectrics and Frequency Control,48(4):1111–1123,July2001.
[10]R.M.Lerner and K.J.Parker.Sonoelasticity images,ultra-
sonic tissue characterization and echographic imaging.In 7th European Communities Workshop,Nijmegen,Nether-lands,1987.
[11]R.M.Lerner,K.J.Parker,J.Holen,R.Gramiak,and R.C.
Waag.Sono-elasticity:medical elasticity images derived from ultrasound signals in mechanically vibrated targets.In Proc.16th int.Symp.Acoustical Imaging,vol.19,pages 317–327,New York,1988.Plenum.
[12]M.O’Donnell,A.R.Skovoroda,B.M.Shapo,and S.Y.
Emelianov.Internal displacement and strain imaging using ultrasonic speckle tracking.IEEE Trans.Ultrason.Ferro-electr.Freq.Control,41:314–325,1994.
[13]J.Ophir.Elastography:ultrasonic imaging of tissue strain
and elastic modulus in vivo.Eur.J.Ultrasound,3:49–70, 1996.
[14]J.Ophir.Scientists use?nite element method in developing
new cancer detection technique.NASA Tech.Briefs,pages 86–87,August1998.
[15]R.W.Prager,R.N.Rohling,A.H.Gee,and L.Berman.
Rapid calibration for3-d freehand ultrasound.Ultrasound in Medicine and Biology,24(6):855–869,1998.
[16]R.N.Rohling,A.H.Gee,and L.Berman.Spatial com-
pounding of3d ultrasound images.Technical Report Tech.
Rep.CUED/F-INFENG/TR270,Univ.of Cambridge,Oc-tober1996.
[17]R.N.Rohling,A.H.Gee,and L.Berman.Automatic reg-
istration of3d ultrasound images.Technical Report Tech.
Rep.CUED/F-INFENG/TR290,Univ.of Cambridge,May 1997.
[18] D.Rubens,M.Hadley,and et al.Sonoelasticity imaging of
prostate cancer:in vitro results.Radiology,195:379–383, 1995.
[19]L.Taylor,B.C.Porter,D.J.Rubens,and K.J.Parker.
Three-dimensional sonoelastography:Principles and prac-tices.Phys.Med.Biol.,2000.
[20]Y.Yamakoshi,J.Sato,and T.Satoa.ultrasonic imaging
of internal vibration of soft tissue under forced vibration.
IEEE Trans.Ultrason.Ferroelectr.Freq.Control,17:45–53,1990.