Ultrasonics 53 (2013) 782–792

Contents lists available at SciVerse ScienceDirect

Ultrasonics

journal homepage: www.elsevier .com/ locate/ul t ras

Comparison of cardiac displacement and strain imaging using ultrasoundradiofrequency and envelope signals

Chi Ma, Tomy Varghese ⇑Department of Medical Physics, University of Wisconsin – Madison, Madison, WI 53705, United States

a r t i c l e i n f o

Article history:Received 13 September 2012Received in revised form 29 October 2012Accepted 2 November 2012Available online 29 November 2012

Keywords:UltrasoundElastographyEchocardiographic strain imagingDeformation imagingCardiac strain imaging

0041-624X/$ - see front matter � 2012 Elsevier B.V.http://dx.doi.org/10.1016/j.ultras.2012.11.005

⇑ Corresponding author. Address: Department of M1111 Highland Ave., Madison, WI 53705, United State+1 608 262 2413.

E-mail addresses: cma8@wisc.edu (C. Ma), tvarghe

a b s t r a c t

Echocardiographic strain imaging is a promising new method for quantifying and displaying the health ofcardiac muscle. Accurate regional myocardial function analysis requires high spatial and temporal reso-lution in addition to fidelity to the underlying deformation. However, all current clinical approaches usespeckle-tracking algorithms applied to B-mode images derived from envelope signals. Such approachesare inherently of lower spatial resolution, since they require larger data blocks for deformation trackingdue to the absence of phase information. In this paper, we compare the strain estimation performanceusing B-mode, envelope and radiofrequency signals, utilizing data acquired from a uniformly elastic tis-sue mimicking phantom, cardiac simulation, and clinical in vivo data. Signal-to-noise ratio improvementsusing radiofrequency signals for linear and phased array geometries were 5.80 dB and 9.48 dB over thatobtained with envelope signals (at peak strain) in phantom studies, respectively. Cardiac simulation stud-ies demonstrate that when averaged over the two cardiac cycles, the mean standard deviation of esti-mated strain using envelope signals from two of the six segments for a short-axes view (anterior andanterolateral) were 48% and 44% higher than that obtained using radiofrequency signals. These segmentswere chosen since one was along while the other was situated lateral to the beam propagation direction.In a similar manner, in vivo analysis on a volunteer also indicate that the standard deviation of the esti-mated strain using B-mode and envelope signals were 16% and 42% higher than that obtained usingradiofrequency signals in the anteroseptal segment, and 45% and 27% in the anterior segment. Theseresults demonstrate the significant reduction in the variability of strain estimated along with improve-ments in the spatial resolution and signal-to-noise ratios obtained using radiofrequency signals.

� 2012 Elsevier B.V. All rights reserved.

1. Introduction

Ultrasound elastography, is a novel and emerging imagingmodality used for depicting tissue elastic properties estimatedusing ultrasound radiofrequency (RF), envelope or B-mode echosignals to track tissue deformations [1–5]. Local displacement andstrain information as surrogates of tissue stiffness are calculatedby tracking local tissue deformation using normalized cross-correlation analysis of the pre- and post-deformation ultrasounddata. Steady state quasi-static elastography can be classified basedon the tissue deformation applied, into three distinct categories [6].Steady-state quasi static deformation is generally used for evalua-tion of variations in tissue stiffness in superficial organs. Applica-tions include detection of breast mass pathologies includingdifferentiation of benign from malignant masses [7–9] and assess-ments of lymph node stiffness [10]. Low frequency (1–10 Hz)

All rights reserved.

edical Physics, 1159 WIMR,s. Tel.: +1 608 265 8797; fax:

se@wisc.edu (T. Varghese).

deformations comprise the second category. These are generatedto perturb tissue along with approaches to track the resulting defor-mation [8]. The third category utilizes quasi static deformationsgenerated by physiological excitation, such as respiratory motion,cardiac muscle deformations and cardiac generated vascular defor-mations. Strain imaging performed for detection of carotid arteryplaques [11,12] and cardiac strain imaging are examples for thiscategory.

Most of the peer-reviewed literature for strain imaging describethe use of linear array transducers, where displacement and strainestimations are generally performed using one-dimensional (1D)[1,2], two dimensional (2D) [3,13–15], and even three-dimensional(3D) [16,17] deformation tracking. 1D approaches track and esti-mate deformations along the beam propagation direction, using agated data segment or window on the pre and post-deformationecho signals. These methods are computationally efficient; how-ever, lateral deformation information is not estimated [1,2]. 2Ddeformation tracking algorithms, utilize small 2D data blocks alsocalled kernels to track tissue deformations and were initially devel-oped for linear array transducers. Many of the shortcomings withthe technique, such as coarse lateral deformation information,

C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792 783

computation time and poor performance in cases where the dis-placement fields were not continuous have been addressed. Konof-agou and Ophir [3] performed computationally intensive weightedinterpolation between neighboring RF A-lines, to obtain higherprecision lateral displacement estimates that was not limited bythe pitch of the transducer used. Zhu and Hall [14] used a modifiedblock-matching algorithm to reduce the search range in 2D block-matching, to improve computational speeds for real-time imple-mentation. In their method, displacement estimates obtained fromthe previous row were used to predict displacements in lowerrows, thereby reducing the search range and increasing computa-tional efficiency. Shi and Varghese [13] successfully estimateddiscontinuous displacement fields utilizing a 2D multi-level algo-rithm. In their algorithm, a pyramidal search scheme was utilizedwhere sub-sampled envelope signals were used in the upper levelsof the pyramid while RF signals were used in the lowest level of thepyramid for deformation tracking. Chen et al. [15] developed aquality guided algorithm for ultrasound displacement trackingover geometrically irregular and disjointed regions, that wasguided by the value of the normalized cross-correlation coefficient.

In general, for deformation tracking and strain estimation usingultrasound echo signals, signal decorrelation due to tissue defor-mation within the processing window or kernel due to changesin both signal amplitude and phase between the pre and post-deformation signal has to be considered [18–21]. As the windowlength or kernel increases in dimension, signal decorrelation dueto the deformation increases concomitantly within the processingwindow or kernel [5]. RF echo signals mapping larger applieddeformation or strain decorrelate faster when compared to smallerapplied strain or deformation. The decorrelation rate is lower withenvelope signals since only amplitude information is present inthese signals [22,23]. Therefore, the deformation information con-tained within similar sized data segments is lower with envelopewhen compared to RF signals. This enables the use of smaller win-dows or data segments when RF echo signals are utilized. As thewindow become smaller, the information contained with onlyenvelope signals become ambiguous for accurate tracking of theunderlying deformations. On the other hand, since RF signals con-tain both amplitude and phase information accurate deformationtracking is possible with reduced decorrelation for smaller windowlengths, thereby improving the spatial resolution for strain ordeformation imaging [24].

Cardiac strain imaging [25–29], for the analysis of regional andglobal cardiac function, has been rapidly developing into a clinicaldiagnostic modality. Current clinical estimations of cardiac regio-nal strain are performed using speckle-tracking algorithms appliedto B-mode signals [25]. B-mode signals that are generated fromenvelope signals are inherently limited in spatial resolution whencompared to RF signals, which also include phase information. Var-ghese and Ophir [23] has previously discussed tradeoffs associatedbetween RF and envelope signals using the ‘strain filter’ concept[30]. The ‘strain filter’ provides a theoretical framework for predict-ing the performance of strain estimation algorithms, by plottingvariations in the elastographic signal-to-noise ratios (SNRe) withapplied strain. In their theoretical and simulation results, theyshow that while SNRe obtained using RF signals is significantlyhigher than envelope signal for low applied strains, the trackingability deteriorates for larger applied strain when compared toenvelope signals due to signal decorrelation. Low strain regionsin the ‘strain filter’ refer to cases where signal decorrelation canstill be modeled using the Cramer Rao lower bound for partiallycorrelated signals [31] and amplitude and phase information arepresent in the signals thereby providing high SNRe. High strain re-gions on the other hand refer to signals where only amplitudeinformation is present and phase information is ambiguous dueto wrap around errors in the phase; in these cases displacement

estimation is governed by the Barankin bound [32]. The ‘strain fil-ter’ varies with ultrasound system parameters such as center fre-quency, bandwidth, beamwidth, and processing parameters forstrain imaging such as the window length or kernel dimensions,overlap, ID or 2D deformation tracking among other parameters[30].

Because of the reduced storage requirement and strain estima-tion performance with reduced signal decorrelation at higher ap-plied deformations, envelope or B-mode signals were used incommercial applications of the strain and strain-rate in cardiacspeckle-tracking based imaging [33–35]. Use of B-mode or enve-lope signals, however, require longer data segments or windowsfor 1D processing and larger 2D blocks or kernels to obtain reason-able SNRe in the resulting strain images. While the temporalresolution is determined by the frame rate, improved spatial reso-lutions, require the use of small cross-correlation data segments orkernels. Several studies have shown that RF signals have significantperformance advantages over envelope signal, as also describedabove [18,36,37]. Lopata et al. [36] performed quantitative analysison local displacements estimated using 2D tracking between RFand envelope signals in a linear array geometry, using a simulateduniformly elastic homogeneous phantom. Their results show thatfor the same applied deformation or strain, the root mean squareerror (RMSE) for estimated displacements is lower with RF signals.Idzenga et al. [37] estimated strain using cross-correlation kernelson the order of 0.32 � 1.22 mm with data from linear array trans-ducers obtaining SNRe values of 5.6 dB with RF signals when com-pared to 1.3 dB obtained using envelope signals.

Despite all the reports presented above by various researchers,to our knowledge, no published work has specifically addressedthe performance of RF, envelope and B-mode strain estimationsfor cardiac or echocardiographic strain imaging. Phased arraytransducers with its small footprint are the de facto standard forechocardiography, due to the limited acoustic windows availablefor imaging the heart. Deformation tracking and strain imagingwith phased array transducers are challenging due to the ultra-sound beam being steered over a ±45� or larger imaging or scan-ning planes. Unlike linear array transducers, displacement andstrain images generated using a phased array transducer have lim-ited lateral resolution at deeper depths, since the beams diverge.Most of the deformation tracking and strain imaging approachesfor cardiac strain imaging therefore utilize 1D cross-correlationanalysis using data segments on the order of 3 mm [26] or4.6 mm [38] as reported in the literature. Our laboratory has devel-oped a hybrid 2D method for deformation tracking and strain esti-mation [39], which was recently expanded to provide a Lagrangiandescription [40] of the strain using kernel lengths on the order of awavelength in the beam direction and 3-Alines in the lateral direc-tion [39,40]. Linear interpolation is utilized for deeper locations toimprove the spatial localization of the lateral displacements. Spa-tial resolution on the order of a wavelength based on kerneldimensions represent the best spatial resolution reported for car-diac strain imaging thus far in the literature [39,40].

In this paper, we compare the performance of deformationtracking and strain estimation using RF and envelope signals forcardiac strain imaging using RF data acquired from both linearand phased array transducers. Comparisons were performed usinga uniformly elastic tissue-mimicking (TM) phantom, cardiac simu-lation data and in vivo clinical data. Cyclic deformations were ap-plied to a uniformly elastic TM phantom during the acquisitionof RF data. A cardiac simulation model was used to demonstratethe differences in strain estimation using RF and envelope signalsover several cardiac cycles. Finally a comparison using RF, envelopeand system generated B-mode data from a healthy volunteer wasperformed. A Lagrangian deformation tracking [40] scheme previ-ously developed in our laboratory was adapted to estimate regio-

784 C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792

nal Lagrangian strains for performance comparisons obtained withRF, envelope and system generated B-mode echo signals.

2. Materials and methods

2.1. TM phantom study

A cyclic deformation system previously designed in our labora-tory was used to evaluate the strain estimation performance with auniformly elastic TM phantom manufactured in our laboratory byMadsen et al. [41]. The device generates cyclic deformations thatdeform and release the TM phantom placed in the middle of a com-pression platform [42]. The uniformly elastic TM phantom usedwas a cube with dimensions 10 � 10 � 10 cm. The Young’s modu-lus of the TM material was 43.08 ± 1.97 kPa, measured using anEnduraTEC ELF 3220 mechanical testing system. (EnduraTECSystems Group, Minnetonka, MN, USA). During the experiment thephantom was placed on the lower platform; a 3% pre-compressionwas applied using the upper compression plate to ensure propercontact. A cyclic deformation frequency of 0.7 Hz, which is closeto the human heart rate, was applied to the deformation of thelower platform using a DC motor. The ultrasound system usedfor RF data acquisition was a Siemens Acuson S2000 real-time clin-ical scanner (Siemens Ultrasound, Mountain View, CA, USA)equipped with a VFX 9L4 linear-array transducer with a center fre-quency of 6 MHz and a 4V1 phased array transducer operating at4 MHz. The imaging depth was 8 cm with a single transmit focusset at 5 cm for both transducers, and dynamic focusing was usedon receive. The RF data acquisition frame rate was 18 Hz for thelinear array transducer and 36 Hz for the phased array transducerunder our imaging acquisition settings. The phased array trans-ducer had a higher frame rate due to the fewer number of beam-lines within the imaged region. Frame skipping was applied tothe phased array data to match the frame rate of 18 Hz obtainedusing the linear array before further processing. This frame ratehas been proven to be sufficient to track displacement and strainvariations in our experimental setup [42]. The RF signal was digi-tized at a sampling frequency of 40 MHz using the ultrasoundresearch interface on the Siemens system.

2.2. Finite element analysis (FEA) based cardiac simulation

A 3D canine heart model [43,44] that was previously incorpo-rated into an ultrasound simulation program in our laboratory[45] was used to generate RF data with cardiac deformation. RFdata over two cardiac cycles of tissue deformation informationwas obtained using this canine heart model with a temporal sam-pling rate of 250 frames/s. Over 1 million scatterers were randomlydistributed in the canine heart model to ensure Rayleigh statisticsfor the generation of the simulated ultrasound RF data. In thisstudy, a 2D mid-cavity slice of the left ventricle along the short axisview was selected and RF data simulated for this view.

2.3. Ultrasound in vivo data acquisition and strain estimation

Under a protocol approved by the UW-Madison Health SciencesInstitutional review board, in vivo data on a volunteer used in thisstudy was acquired at University of Wisconsin-Madison Adult Car-diology Clinics. Informed consent was obtained prior to ultrasoundscanning. The volunteer was scanned using a GE Vivid 7 system(GE Ultrasound, Waukesha, WI, USA) with a 2.5 MHz phased arraytransducer. RF signals along a parasternal short axis view atmid-cavity at a frame rate of 34.1 fps was acquired at a 20 MHzsampling rate.

2.4. 2D processing for displacement tracking and strain estimation

Envelope signals were generated from the RF data using Hilberttransformation [46,47]. The resulting RF and envelope signals werethen processed using a multi-level speckle tracking method [39] togenerate the local displacement field along the beam propagationdirection. For the in vivo data on the volunteer the B-mode cineloop directly generated from the ultrasound system was utilized.Our displacement tracking and strain imaging algorithm clearlyindentifies the location of the pre- and post- deformation signalsover the data loop. Four levels of tracking were performed with75% window overlaps at each level. 2D cross-correlation processingkernels on the order of a single wavelength along the beam direc-tion and three A-lines in the lateral direction were used for thefinal processing kernel dimensions unless specified otherwise. A9-point least squares linear fit of the displacements was then usedto compute the local strain field.

Regional accumulated displacements and strains for the cardiacsimulation and in vivo data set were then calculated using ourLagrangian tracking method [40], which enables tracking of tissuedeformation around a point/region in myocardial tissue as itmoves/traverses through space and time. We have previously dem-onstrated that Lagrangian tracking outperforms traditional Euleri-an tracking where deformations around a fixed spatial coordinateare tracked through time [40].

Although scanning was performed with different transducercenter frequencies for the TM phantom, and in vivo data respec-tively due to the different imaging depths, all the comparisons be-tween RF, envelope and B-mode signals were performed on thesame exact data set. In addition, for data processing the final kerneldimension is always specified in terms of the ultrasound wave-length and not based on the center frequency. The performancecomparison between RF, envelope and B-mode signals was the fo-cus of this paper.

3. Results

3.1. Uniformly elastic TM phantom results

Plots of the mean displacement estimates along the beam direc-tion and within a small region of interest (ROI) at the center of theuniformly elastic TM phantom are shown in Fig. 1. A 1 cm2 ROI wasselected at the center of the phantom at a 5 cm depth, where thefocus of the transducer was placed. The mean and standard devia-tion of the local displacements estimated were computed over thisROI for 10 independent cyclic deformations of the phantom in thecompression apparatus described in the previous section.Fig. 1(a.1) presents the variations in the mean displacement esti-mated over 10 independent data sets acquired using a linear arraytransducer for both RF and envelope signals. Error bars on the plotsdenote the variation in the mean displacement over the indepen-dent realizations. Both the curves are similar and no significant dif-ferences are observed between the mean displacements estimatedusing the linear array transducer. Fig. 1(a.2) shows the variation inthe standard deviation of the displacements estimated, with theerror bars denoting its variation over the independent realizationsrespectively.

Fig. 1(b.1) and (b.2) depict similar analysis of the mean andstandard deviation of the local displacements estimated using aphased array transducer. Again, a 1 cm2 ROI was selected at thecenter of the phantom at a 5 cm depth. Fig. 1(b.1) presents the vari-ations in the mean displacement estimated over 10 independentdata sets acquired using a phased array transducer for both RFand envelope signals. Error bars on the plots denote the variationin the mean displacement over the independent realizations.

Fig. 1. Mean displacement and standard deviation estimates obtained using a cyclic deformation of a uniformly elastic TM phantom study using (a) linear array transducer,(b) phased array transducer. Plots a.1 and b.1 depict mean displacement variation within the ROI, while plots a.2 and b.2 show the mean standard deviation of displacementestimates. Error bars denote the standard deviation of the mean displacement/strain/standard deviation over 10 independent realizations.

C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792 785

Fig. 1(b.2) shows the variation in the standard deviation of the dis-placements estimated, with error bars denoting its variation overthe independent realizations respectively. Observe that, while themean displacement variations remain indistinguishable betweenRF and envelope signals, the peak standard deviation obtainedusing envelope signals is 96% higher than that obtained using RFsignals, indicating a more noisy estimation of the displacementsfrom envelope signal in the phased array setup.

Strain variations along the beam direction for the TM phantomstudy are shown in Fig. 2. Fig. 2(a.1) presents the mean strain over10 independent realizations within the ROI with data acquiredfrom linear array transducer. Fig. 2(a.2) shows the standard devia-tion of strain distribution over the independent realizations. Errorbars denote the variations of mean strain and standard deviationover the 10 independent realizations in these figures. Fig. 2(a.3)presents plots of the mean SNRe over 10 independent realizations,where SNRe is defined as previously described in [30]:

SNRe ¼S1

r1ð1Þ

where S1 denotes the mean strains over the ROI and r1 is the corre-sponding standard deviation. Error bars denotes the variations ofmean SNRe over the 10 independent realizations. Note that we areable to plot the SNRe since we utilize a uniformly elastic phantom,where the strain distribution would be uniform for a specified ap-plied deformation. Fig. 2(a.1) shows that estimated mean strainsare very similar for both RF and envelope processing. However,the mean values of the standard deviation are significantly largerfor strains estimated using envelope signals when compared tothat obtained using RF signals over the entire cyclic deformation.

Results obtained using a phased array transducer are shown inFig. 2(b.1)–(b.3). In Fig. 2(b.1), for the peak strain values, envelopeprocessing fails to depict the sinusoidal shape of the deformationgenerated by the compression system. The variation in the meanstandard deviation observed between envelope and RF signal arehigher than that seen for the linear array transducer, which leadsto the significant SNRe difference between RF and envelope resultsfor the phased array data.

Fig. 3 provides an overview of variations in the mean SNRe val-ues for different lengths of the final processing kernel along thebeam direction used in the multi-level speckle tracking method.Comparisons were made between RF and envelope signals. MeanSNRe values were obtained at the peak strain value and computedover 10 independent realizations. Error bars denotes the variationsin the mean SNRe over the 10 independent realizations. Fig. 3a andb represent results from linear array and phased array transducersrespectively. SNRe values for RF signals are significantly higherthan envelope signals in both plots. For the final processing ker-nel’s dimension of 1 wavelength along the beam direction, theSNRe improvement using RF signals for linear and phased arrayswas 5.80 dB and 9.48 dB respectively, when compared with enve-lope signals.

3.2. Simulation results

Segmental mean accumulated displacement variations overtime estimated from the cardiac FEA model and subsequentultrasound simulations obtained using Lagrangian tracking areshown in Fig. 4(a.1) and (b.1) for two of the segments. The left ven-tricular wall was segmented based on the standard proposed by

Fig. 2. Variations in the mean strain and standard deviation obtained using a cyclic deformation of a uniformly elastic TM phantom using linear array (a.1–a.3) and phasedarray transducers (b.1–b.3). Plots a.1 and b.1 depict mean strain variation within the ROI while a.2 and b.2 plot the variation in the mean standard deviation of the strain. Plotsa.3 and b.3 show the mean SNRe for the two transducer geometries. Error bars denote the standard deviation of the mean displacement/strain/standard deviation over 10independent realizations.

786 C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792

the American Society of Echocardiography [48]. Comparisons be-tween RF and envelope results are shown over two cardiac cyclesfor the anterior and anterolateral segment, plotted separately inFig. 4(a.1) and (b.1). These segments were selected since in thesimulation setup, the anterior part of left ventricular wall mainlymoves parallel to the insonification direction whereas the antero-lateral segment incurs more perpendicular or lateral deformationwith respect to the insonification direction. This is observed whencomparing Fig. 4(a.1) with (b.1), where the peak amplitude ofmean displacement in the anterior segment (2.25 mm) is higherthan that of anterolateral segment (0.42 mm). Similar to the resultobtained using the phantom study using the linear array trans-ducer, Fig. 4(a.1) and (a.2) show that mean displacement curvesgenerated by RF and envelope signals are comparable, and both

curves closely follow the mean displacement provided by theFEA model solution. Percent differences between RF, envelopeand FEA results are calculated for the mean displacement curvesover two cardiac cycles, as shown in Table 1. Absolute values ofthe displacement were utilized to determine the error betweenRF and envelope processing when compared to the FEA result. Ob-serve that in the anterior segment; the difference between enve-lope and RF results was 2.8%, while they deviate from the FEAresults by �3.8% for the envelope and 2.8% with RF processingrespectively. For the anterolateral segment, these differences were0.8%, �1.9% and �2.7% respectively. The standard deviation of theaccumulated displacement estimates over time with each segmentare plotted in Fig. 4(a.2) and (b.2) for the anterior and anterolateralsegments respectively. The variation in the standard deviation

Fig. 3. Variations of the mean SNRe at the peak strain value with respect to the final processing kernel’s axial dimension over 10 independent realizations for (a) linear arrayand (b) phased array transducers. Comparisons are made between RF and envelope signals.

Fig. 4. Comparison of the mean segmental accumulated displacement variation (left) and its standard deviation (right) with time for RF, envelope and FEA results. Plots (a)and (b) correspond to results from anterior and anterolateral segment respectively.

C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792 787

plots for both RF and envelope processing closely follow FEAresults. The percent difference between envelope and RF resultsfor the anterior segment was 2.1%, while they deviate from FEAresults by �4.6% using envelope and 2.4% using RF signals respec-tively. For the anterolateral segment, these differences were 0.8%,2.2% and 1.5% respectively.

Differences in the segmental mean accumulated strains areshown in Fig. 5(a.1) and (b.1). Similar to the results shown withthe phantom study, mean strains generated from RF and envelopedata are similar and follow the ideal FEA generated strains. Asshown in Table 1, the percent difference between envelope andRF based strain estimation results in the anterior segment was

�14%. The envelope results deviate from FEA results by 11% and29% with the RF estimations respectively. In the anterolateral seg-ment, these differences were 4.8%, 2.4% and �2.3% respectively.Absolute strain estimates were utilized for the difference calcula-tions. The standard deviation of the estimated RF strains followthe standard deviation of the 3D canine FEA model, the envelopeestimates on the other hand deviate significantly from the idealFEA curve, as shown in Fig. 5(a.2) and (b.2). Observe that after aver-aging over the two cardiac cycles, the percent difference betweenenvelope and RF results in the anterior segment was 48%. Note thatthe mean deviation from FEA results was 212% using envelope sig-nals and 110% for the RF data respectively. For the anterolateral

Table 1Percent differences between RF, envelope and FEA results for the cardiac simulation study over two cardiac cycles. Absolute values were used for displacement and strainestimates for computation of the percent difference.

Percent difference

Mean displacement (%) Displacement standard deviation (%) Mean strain (%) Strain standard deviation (%)

Anterior segmentRF vs. FEA �6.4 2.4 29 110Envelope vs. FEA �3.8 4.6 11 212Envelope vs. RF 2.8 2.1 �14 48

Anterolateral segmentRF vs. FEA �2.7 1.5 �2.3 5.3Envelope vs. FEA �1.9 2.2 2.4 51Envelope vs. RF 0.8 0.8 4.8 44

Fig. 5. Comparison of the mean segmental accumulated strain variation (left) and its standard deviation (right) with time for RF, envelope and FEA results. Plots (a) and (b)correspond to results from anterior and anterolateral segment respectively.

788 C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792

segment, these percent differences were 44%, 51% and 5.3% respec-tively. Note that SNRe plots are not presented for the cardiac sim-ulations since cardiac tissue is anisotropic and the deformationsgenerated are complex with translation, rotation and torsionaldeformations all present during a cardiac cycle.

3.3. In vivo results

Fig. 6 provides a representative example of the accumulateddisplacement and strain mapping at the end-systolic phase of thecardiac cycle. Only the regions within the left ventricular wall con-tours are shown overlaid on their corresponding B-mode images.Due to the noisy nature of lateral displacements and strains gener-ated with a phased array setup, displacements and strainsdisplayed were estimated along each A-line. Fig. 6a–c representresults from RF, envelope and system generated B-mode signalsrespectively. Images along the left column are displacement maps,

and images on the right column are strain maps. Both positive(red) and negative (blue) strain information (with respect to theultrasound beam propagation direction) is shown indicating bothcontractile, where tissue sections move closer together depictedas red regions, and stretching of the cardiac muscle as blue regions.For displacement mapping, the inferior side of myocardium pro-vide positive displacement values while the anterior side presentwith negative displacement values, indicating that the myocar-dium is undergoing a counter clock-wise twist at end-systole. Neg-ative strain values along the anteroseptal, inferolateral and inferiorsegments for strain mapping indicate contraction in those regionsat end-systole.

In-vivo comparisons of the mean segmental accumulated dis-placement variations over time between RF, envelope and systemgenerated B-mode signals for a volunteer are shown in Fig. 7(a.1)and (b.1), for the anteroseptal and anterior segments. Using thefact that, in a healthy heart, myocardial walls will return to its ori-

Fig. 6. Accumulated displacement (left) and strain (right) images acquired in vivo for a volunteer at the mid-systolic phase of a cardiac cycle superimposed on the B-modeimage. From (a) to (c), images represent results using (a) RF, (b) envelope and (c) system generated B-mode results. Displacement in the insonification direction towards thetransducer is negative (blue) while displacement away from the transducer is positive (red) with respect to the beam propagation direction. (For interpretation of thereferences to colour in this figure legend, the reader is referred to the web version of this article.)

C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792 789

ginal status after each cardiac cycle, the lowest values in the curveswere set to 0 to ensure better visual comparison. The standarddeviation of the segmental accumulated displacements are pre-sented in Fig. 7(a.2) and (b.2). Percent difference calculations be-tween B-mode, envelope and RF results are shown in Table 2.Note that, when averaged over two cardiac cycles, the standarddeviation of the estimated displacement from anteroseptal andanterior segment for B-mode signals were 18% and 54% higher than

that obtained for corresponding RF signal estimations. For boththese segments the differences were 11% between envelope andRF signals.

In a similar manner, mean segmental accumulated strain varia-tions over time between RF, envelope and system generatedB-mode signals for a volunteer are shown in Fig. 8(a.1) and (b.1),for the anteroseptal and anterior segments. Standard deviation ofthe segmental accumulated strains are shown in Fig. 8(a.2) and

Fig. 7. Comparison of the mean segmental accumulated displacement variation and its standard deviation with time for RF, envelope and system generated B-mode results.Plots (a) and (b) represent results from the anteroseptal and anterior segments respectively.

Table 2Percent differences between RF, envelope and B-mode results using mean standarddeviation values estimated over 2 cardiac cycles. Mean displacements and strainswere not compared since an actual estimate is not available for in vivo data.

Percent difference

Displacement standarddeviation (%)

Strain standarddeviation (%)

Anteroseptal segmentB-mode vs. RF 18 16Envelope vs. RF 11 42

Anterior segmentB-mode vs. RF 54 45Envelope vs. RF 11 27

790 C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792

(b.2), while quantitative comparison results are shown in Table 2.Over two cardiac cycles, the standard deviation of strain estimatedfrom B-mode signals from the anteroseptal and anterior segmentare 16% and 45% higher than that obtained with RF signal estima-tions. Comparison between envelope signals for these segmentsindicated a 42% and 27% difference when compared to strain esti-mated using RF signals.

4. Discussion

In this paper we compare the performance of displacementtracking and strain estimation using RF and envelope signals forcardiac or echocardiographic strain imaging. The comparisonswere done using a uniformly elastic TM phantom, cardiac simula-tion data and in vivo clinical data. We utilized a final cross-correla-tion kernel length of one wavelength along the beam direction for

processing using RF, envelope and system generated B-mode sig-nals. The purpose of such a small processing kernel is to ensurehigh spatial resolution in the estimated displacement and strainimages. The pre- and post-deformation data segments were posi-tioned using our multi-level hybrid 2D algorithm [39], and the per-formance comparison directly reflects the fact that displacementand strain estimation was obtained with correctly aligned signals.This alignment is crucial and may not be achieved with other algo-rithms that do not make this additional processing a priority.

Fig. 1 indicates comparable displacement estimation perfor-mance using linear arrays between RF and envelope signals. Forthe phased array transducer, the variation in the standard devia-tion of the displacements estimated is higher (96%) for envelopewhen compared to RF signals. Fig. 2 suggests that strains estimatedusing RF signals provide improved SNRe performance when com-pared to envelope signals for linear (5.8 dB) as well as phased arraytransducers (9.5 dB). Our results indicate a higher variability inenvelope strain estimation when a small cross-correlation kernelis used.

For a comparison of SNRe improvement at different spatial res-olutions, only the final processing kernel’s dimension was varied inFig. 3. The dimensions of the kernel in the first three levels of track-ing kernels, the final processing kernel’s lateral dimension, windowoverlap percentage and least squares fit length method remainedunchanged. When the processing kernel’s dimension along thebeam propagation direction is reduced to a half wavelength, theSNRe obtained with envelope signals are lower than 0 dB for bothlinear and phased array transducers, suggesting that the estimatedstrain signal and noise signals are indistinguishable.

Results from the cardiac simulation study demonstrate similarresults as described in the uniformly elastic TM phantom study,

Fig. 8. Comparison of the mean segmental accumulated strain variation and its standard deviation with time for RF, envelope and system generated B-mode results. Plots (a)and (b) represent results from the anteroseptal and anterior segments respectively.

C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792 791

as shown in Figs. 4 and 5. The regional mean accumulated strainare similar, however, the mean standard deviation is significantlyhigher (48% and 44% from anteroseptal and anterior segment) forstrain estimated using envelope signals when compared to RF sig-nals, indicating increased spatial variability in the estimates. Thisresult suggests that strain estimated using envelope signals mayyield more unstable results than that estimated using RF echosignals.

The clinical study included three different data formats insteadof the two evaluated using TM phantom and simulation studies.We also utilized the system generated B-mode images to evaluatesimilarities and differences between RF and RF generated envelopesignals. As illustrated in Table 2, displacement and strain estimatedfrom the system generated B-mode and envelope signals indicatelarger percent differences in the mean standard deviation plotswhen compared to that obtained using RF echo signals.

5. Conclusion

In this paper, we illustrate the performance advantage obtainedwith utilizing RF echo signals over envelope or B-mode signals forcardiac or echocardiographic strain imaging. This is particularlytrue when small speckle tracking kernels are utilized to improvespatial resolution in the strain images. Temporal resolution isdetermined by the frame rate of the system. Two performancemetrics, namely the regional mean displacement/strain and thestandard deviation of the regional mean displacement/strain wereevaluated in this paper. In phantom studies, the SNRe improvementobtained using RF signals for linear and phased array geometrieswas 5.80 dB and 9.48 dB respectively, when compared to that ob-tained with envelope signals at the peak strain value. Results ob-

tained from the cardiac simulation study also indicate that whenaveraged over two cardiac cycles, the standard deviation of esti-mated strain using envelope signals from the anterior and antero-lateral segment are 48% and 44% higher than that obtained usingRF signals. In vivo analysis on a volunteer shows that the standarddeviation of the estimated strain using B-mode and envelope sig-nals were 16% and 42% higher than that obtained using RF signalsin the anteroseptal segment, and 45% and 27% in the anteriorsegment.

Acknowledgements

This work is supported in part by NIH grant 5R21EB010098-02and R01 CA112192-05. The authors gratefully acknowledge the useof the cardiac mechanics model from the Cardiac Mechanics Re-search Group at UCSD.

References

[1] J. Ophir, I. Cespedes, H. Ponnekanti, Y. Yazdi, X. Li, Elastography: a quantitativemethod for imaging the elasticity of biological tissues, Ultrason Imaging 13 (2)(1991) 111–134.

[2] I. Cespedes, J. Ophir, H. Ponnekanti, N. Maklad, Elastography: elasticity imagingusing ultrasound with application to muscle and breast in vivo, UltrasonImaging 15 (2) (1993) 73–88.

[3] E. Konofagou, J. Ophir, A new elastographic method for estimation and imagingof lateral displacements, lateral strains, corrected axial strains and poisson’sratios in tissues, Ultrasound Med. Biol. 24 (8) (1998) 1183–1199.

[4] F. Kallel, J. Ophir, A least-squares strain estimator for elastography, UltrasonImaging 19 (3) (1997) 195–208.

[5] T. Varghese, M. Bilgen, J. Ophir, Multiresolution imaging in elastography, IEEETrans. Ultrason Ferroelectr. Freq. Control 45 (1) (1998) 65–75.

[6] T. Varghese, Quasi-static ultrasound elastography, Ultrasound Clin. 4 (3)(2009) 323–338.

792 C. Ma, T. Varghese / Ultrasonics 53 (2013) 782–792

[7] B.S. Garra, E.I. Cespedes, J. Ophir, S.R. Spratt, R.A. Zuurbier, C.M. Magnant, M.F.Pennanen, Elastography of breast lesions: initial clinical results, Radiology 202(1) (1997) 79–86.

[8] T.J. Hall, Y. Zhu, C.S. Spalding, In vivo real-time freehand palpation imaging,Ultrasound Med. Biol. 29 (3) (2003) 427–435.

[9] H. Xu, M. Rao, T. Varghese, A. Sommer, S. Baker, T.J. Hall, G.A. Sisney, E.S.Burnside, Axial-shear strain imaging for differentiating benign and malignantbreast masses, Ultrasound Med. Biol. 36 (11) (2010) 1813–1824.

[10] F. Alam, K. Naito, J. Horiguchi, H. Fukuda, T. Tachikake, K. Ito, Accuracy ofsonographic elastography in the differential diagnosis of enlarged cervicallymph nodes: comparison with conventional B-mode sonography, AJR Am. J.Roentgenol. 191 (2) (2008) 604–610.

[11] H. Shi, T. Varghese, C.C. Mitchell, M. McCormick, R.J. Dempsey, M.A. Kliewer, Invivo attenuation and equivalent scatterer size parameters for atheroscleroticcarotid plaque: preliminary results, Ultrasonics 49 (8) (2009) 779–785.

[12] H. Shi, C.C. Mitchell, M. McCormick, M.A. Kliewer, R.J. Dempsey, T. Varghese,Preliminary in vivo atherosclerotic carotid plaque characterization using theaccumulated axial strain and relative lateral shift strain indices, Phys. Med.Biol. 53 (22) (2008) 6377–6394.

[13] H.R. Shi, T. Varghese, Two-dimensional multi-level strain estimation fordiscontinuous tissue, Phys. Med. Biol. 52 (2) (2007) 389–401.

[14] Y. Zhu, T.J. Hall, A modified block matching method for real-time freehandstrain imaging, Ultrason Imaging 24 (3) (2002) 161–176.

[15] L. Chen, G.M. Treece, J.E. Lindop, A.H. Gee, R.W. Prager, A quality-guideddisplacement tracking algorithm for ultrasonic elasticity imaging, Med. ImageAnal. 13 (2) (2009) 286–296.

[16] X. Chen, H. Xie, R. Erkamp, K. Kim, C. Jia, J.M. Rubin, M. O’Donnell, 3-Dcorrelation-based speckle tracking, Ultrason Imaging 27 (1) (2005) 21–36.

[17] A. Elen, H.F. Choi, D. Loeckx, H. Gao, P. Claus, P. Suetens, F. Maes, J. D’Hooge,Three-dimensional cardiac strain estimation using spatio-temporal elasticregistration of ultrasound images: a feasibility study, IEEE Trans. Med. Imaging27 (11) (2008) 1580–1591.

[18] S.K. Alam, J. Ophir, On the use of envelope and rf signal decorrelation as tissuestrain estimators, Ultrasound Med. Biol. 23 (9) (1997) 1427–1433.

[19] I. Cespedes, M. Insana, J. Ophir, Theoretical bounds on strain estimation inelastography, IEEE Trans. Ultrason., Ferroelectr. Freq. Control 42 (5) (1995)969–972.

[20] T. Varghese, J. Ophir, E. Konofagou, F. Kallel, R. Righetti, Tradeoffs inelastographic imaging, Ultrasonic Imaging 23 (4) (2001) 216–248.

[21] T. Varghese, J. Ophir, I. Cespedes, Noise reduction in elastograms usingtemporal stretching with multicompression averaging, Ultrasound Med. Biol.22 (8) (1996) 1043–1052.

[22] S.K. Alam, J. Ophir, On the use of envelope and RF signal decorrelation as tissuestrain estimators, Ultrasound Med. Biol. 23 (9) (1997) 1427–1433.

[23] T. Varghese, J. Ophir, Characterization of elastographic noise using theenvelope of echo signals, Ultrasound Med. Biol. 24 (4) (1998) 543–555.

[24] S.K. Alam, J. Ophir, T. Varghese, Elastographic axial resolution criteria: anexperimental study, IEEE Trans. Ultrason Ferroelectr. Freq. Control 47 (1)(2000) 304–309.

[25] J.R. Gorcsan, H. Tanaka, Echocardiographic assessment of myocardial strain, J.Am. Coll. Cardiol. 58 (14) (2011) 1401–1413.

[26] T. Varghese, J.A. Zagzebski, P. Rahko, C.S. Breburda, Ultrasonic imaging ofmyocardial strain using cardiac elastography, Ultrason Imaging 25 (1) (2003)1–16.

[27] E.E. Konofagou, J. D’Hooge, J. Ophir, Myocardial elastography – a feasibilitystudy in vivo, Ultrasound Med. Biol. 28 (4) (2002) 475–482.

[28] E. Konofagou, W. Lee, C. Ingrassia, A theoretical performance assessment toolfor myocardial elastography, Conf. Proc. IEEE Eng. Med. Biol. Soc. 1 (2005) 985–988.

[29] J. D’Hooge, A. Heimdal, F. Jamal, T. Kukulski, B. Bijnens, F. Rademakers, L. Hatle,P. Suetens, G.R. Sutherland, Regional strain and strain rate measurements bycardiac ultrasound: principles, implementation and limitations, Eur. J.Echocardiogr. 1 (3) (2000) 154–170.

[30] T. Varghese, J. Ophir, A theoretical framework for performancecharacterization of elastography: the strain filter, IEEE Trans. UltrasonFerroelectr. Freq. Control 44 (1) (1997) 164–172.

[31] W.F. Walker, G.E. Trahey, A fundamental limit on delay estimation usingpartially correlated speckle signals, IEEE Trans. Ultrason Ferroelectr. Freq.Control 42 (2) (1995) 301–308.

[32] E. Weinstein, A. Weiss, Fundamental limitations in passive time delayestimation Part II: wide-band systems, IEEE Trans. Acoust Speech SignalProc. 31 (1984) 1064–1078.

[33] M. Ashraf, X.K. Li, M.T. Young, A.J. Jensen, J. Pemberton, L. Hui, P. Lysyansky, Z.Friedman, B. Park, D.J. Sahn, Delineation of cardiac twist by a sonographicallybased 2-dimensional strain analysis method: an in vitro validation study, J.Ultrasound Med. 25 (9) (2006) 1193–1198.

[34] A. Manovel, D. Dawson, B. Smith, P. Nihoyannopoulos, Assessment of leftventricular function by different speckle-tracking software, Eur. J.Echocardiogr. 11 (5) (2010) 417–421.

[35] E. Grabskaya, C. Spira, R. Hoffmann, E. Altiok, C. Ocklenburg, M. Becker,Myocardial rotation but not circumferential strain is transducer angledependent: a speckle tracking echocardiography study, Echocardiography 27(7) (2010) 809–814.

[36] R.G.P. Lopata, M.M. Nillesen, H.H.G. Hansen, I.H. Gerrits, J.M. Thijssen, C.L. deKorte, Performance evaluation of methods for two-dimensional displacementand strain estimation using ultrasound radio frequency data, Ultrasound Med.Biol. 35 (5) (2009) 796–812.

[37] T. Idzenga, H. Hansen, R. Lopata, C. de Korte, Estimation of Longitudinal shearstrain in the carotid arterial wall using ultrasound radiofrequency data,Ultraschall Med, 2011.

[38] S.J. Okrasinski, B. Ramachandran, E.E. Konofagou, Assessment of myocardialelastography performance in phantoms under combined physiologic motionconfigurations with preliminary in vivo feasibility, Phys. Med. Biol. 57 (17)(2012) 5633–5650, http://dx.doi.org/10.1088/0031-9155/57/17/5633.

[39] H. Chen, T. Varghese, Multilevel hybrid 2D strain imaging algorithm forultrasound sector/phased arrays, Med. Phys. 36 (6) (2009) 2098–2106.

[40] C. Ma, T. Varghese, Lagrangian displacement tracking using a polar gridbetween endocardial and epicardial contours for cardiac strain imaging, Med.Phys. 39 (4) (2012) 1779–1792.

[41] E.L. Madsen, M.A. Hobson, H. Shi, T. Varghese, G.R. Frank, Tissue-mimickingagar/gelatin materials for use in heterogeneous elastography phantoms, Phys.Med. Biol. 50 (23) (2005) 5597–5618.

[42] H. Chen, T. Varghese, P.S. Rahko, J.A. Zagzebski, Ultrasound frame raterequirements for cardiac elastography: experimental and in vivo results,Ultrasonics 49 (1) (2009) 98–111.

[43] R. Mazhari, J.H. Omens, L.K. Waldman, A.D. McCulloch, Regional myocardialperfusion and mechanics: a model-based method of analysis, Ann. Biomed.Eng. 26 (5) (1998) 743–755.

[44] A.D. McCulloch, R. Mazhari, Regional myocardial mechanics: integrativecomputational models of flow-function relations, J. Nucl. Cardiol. 8 (4)(2001) 506–519.

[45] H. Chen, T. Varghese, Three-dimensional canine heart model for cardiacelastography, Med. Phys. 37 (11) (2010) 5876–5886.

[46] J.W. Luo, K. Fujikura, S. Homma, E.E. Konofagou, Myocardial elastography atboth high temporal and spatial resolution for the detection of infarcts,Ultrasound Med. Biol. 33 (8) (2007) 1206–1223.

[47] M. Vogt, H. Ermert, Development and evaluation of a high-frequencyultrasound-based system for in vivo strain imaging of the skin, IEEE Trans.Ultrason. Ferroelectr. Freq. Control 52 (3) (2005) 375–385.

[48] M.D. Cerqueira, N.J. Weissman, V. Dilsizian, A.K. Jacobs, S. Kaul, W.K. Laskey,D.J. Pennell, J.A. Rumberger, T. Ryan, M.S. Verani, Standardized myocardialsegmentation and nomenclature for tomographic imaging of the heart. Astatement for healthcare professionals from the cardiac imaging committee ofthe council on clinical cardiology of the American heart association, Int. J.Cardiovasc. Imaging 18 (1) (2002) 539–542.