- Email: [email protected]
Multimode Ultrasound Breast Imaging Using a New Array Transducer Configuration∗
Ultrasound in Med. & Biol., Vol. 36, No. 4, pp. 637–646, 2010 Copyright Ó 2010 World Federation for Ultrasound in Medicine & Biology Printed in the USA. All rights reserved 0301-5629/10/$–see front matter
doi:10.1016/j.ultrasmedbio.2010.01.006
d
Original Contribution MULTIMODE ULTRASOUND BREAST IMAGING USING A NEW ARRAY TRANSDUCER CONFIGURATION* MOK-KUN JEONG and SUNG-JAE KWON Departments of Electronic and Communication Engineering, Daejin University, Pocheon, Gyeonggi, Korea (Received 1 June 2009; revised 22 December 2009; in final form 22 January 2010)
Abstract—This article presents a diagnostic ultrasound imaging technique that can be used in imaging protruding objects such as a human breast using two opposing array transducers. Because two B-mode images obtained from each of the two linear array transducers facing each other represent the same imaging area viewed in different directions, the image quality can be improved using a compounding technique. Using one array as a transmitter and the other as a receiver, the speed of sound distribution in a medium interposed between them is also reconstructed. In addition, because the spacing between the two arrays can be finely controlled, strain image can also be obtained. This new method can be used to produce a compound B-mode image, a speed of sound image, and a strain image of the same region-of-interest, making it possible to obtain more information leading to better diagnosis. Experimental results on a phantom containing a cylinder of different speed of sound and elasticity confirm that the proposed method is useful in obtaining compound and speed of sound images as well as strain images. (E-mail: [email protected]) Ó 2010 World Federation for Ultrasound in Medicine & Biology. Key Words: Attenuation, Backprojection, Breast, Compounding, Reconstruction, Speckle, Speed of sound, Strain imaging.
relatively long data acquisition time and discomfort to patients, it has not yet been put to practical use. To overcome these limitations, several researchers proposed data acquisition methods based on linear array transducers (Huang et al. 2004; Krueger et al. 1996). As with mammography, using a linear array transducer and a plane reflector, the SOS image of a human tissue positioned between them was obtained. Compared with tomographic reconstruction methods that collect data over an angle of 360 , the linear array method acquires data over a limited range of angles, making it difficult to produce a complete artifact-free image (Krueger et al. 1996). Huang et al. (2004) compensated for distortion in SOS image using structural information extracted from B-mode image. To superpose images obtained from multiple viewing angles in full angle spatial compounding, Hansen et al. (2007) reconstructed the SOS distribution by placing a reflector behind an object of interest and using the filtered backprojection of the echo data from the reflector. Jeong et al. (2008) obtained SOS images in transmission mode using two transducers that face each other. Compound imaging techniques acquire data using spatial or frequency diversity and average images of the same region to achieve less speckle noise and better contrast
INTRODUCTION Although mammography is widely used in the diagnosis of breast cancer, it may be subject to potential radiation hazard. Breast cancer has different ultrasound characteristics from normal healthy tissue in terms of speed of sound (SOS) and attenuation (Goss et al. 1978). For the case of diagnosing breasts in medical ultrasound imaging, tissue parameter imaging can be used together with B-mode imaging in a complementary manner. Since the breast protrudes from a human body, it is possible to obtain attenuation and SOS images by using circular transducers or by rotating linear arrays over 360 in transmission or reflection mode using tomographic reconstruction principles (Glover et al. 1977; Greenleaf et al. 1975, 1981). Although this approach enables us to obtain more information about lesion, the use of circular transducers or linear arrays entails the immersion of objects to be imaged in water. Due to
Address correspondence to: Prof. Sung-Jae Kwon, Departments of Electronic and Communication Engineering, Daejin University, Pocheon, Gyeonggi 487-711, Korea. E-mail: [email protected] * Part of this paper was presented at 2008 IEEE Ultrasonics Symposium, Beijing, China, by Jeong et al. under the title ‘‘Ultrasound breast imaging technique using two opposing transducer arrays.’’ 637
638
Ultrasound in Medicine and Biology
(Robinson 1984; Wilhjelm et al. 2000). Carson et al. (1981) introduced a method of obtaining compound image together with SOS image using circular array transducers. Recently, there has been much research into elasticity imaging of the breast. Since tumor or cancer in soft tissue such as the breast and prostate tends to be stiffer than the surrounding tissue, elasticity imaging that visualizes the degree of stiffness is of great help in diagnosing it (Hall 2004; Ophir et al. 1991; Sarvazyan et al. 1998). Human tissues tend to exhibit a higher contrast in elasticity than other tissue parameters. Therefore, the elasticity parameter is amenable to imaging. In this article, we obtain spatial compound, SOS, and strain images using two opposing linear array transducers. B-mode or strain images are obtained independently in reflection mode from each of the two array transducers, and are compounded to improve the image quality. In transmission mode, however, we use an imaging configuration where one array transducer is used as a transmitter and the other is used as a receiver. The time of flight (TOF) is measured and an SOS image is reconstructed. A combination of the B-mode, strain, and SOS imaging can increase the accuracy of lesion diagnosis. This article presents each imaging method and experimental results. SPATIAL COMPOUND IMAGING Compound imaging is used to reduce speckle noise as well as artifacts such as shadowing and reverberation by averaging a certain number of images whose correlations are small. However, the degradation of resolution and signal-to-noise ratio (SNR) with increasing depth cannot be compensated for. Conventional spatial compound B-mode images are produced by averaging individual images of the same imaging region obtained by steering scan lines at different angles. However, the number of useful images obtainable with a single transducer array is limited by its characteristics or the performance of an ultrasound imaging system employing it. If two linear array transducers are arranged such that they face each other, compound images can be reconstructed separately using each of them and those compound images can be averaged again. As a result, the images can be compensated for their depth-dependent characteristics. Figure 1 shows the imaging configuration used in this article where the two linear array transducers face each other. A slanted imaging area that makes an angle of q to the upper transducer array normal by steering ultrasound beams is also indicated in dashed lines in the figure. SOS IMAGING The measurement of ultrasound speed has been a topic of interest since it can aid in tissue characterization, beamforming, scan conversion, and image registration
Volume 36, Number 4, 2010
Fig. 1. Compounding of independently obtained images using each of two opposing array transducers.
(Duric et al. 2008; Glover et al. 1977; Goss et al. 1978; Greenleaf et al. 1975, 1981; Hansen et al. 2007; Huang et al. 2004; Krueger et al. 1996; Kru¨cker et al. 2004; Li 2008, 2009; Robinson et al. 1991). Recently, the work of Duric et al. has been quite noteworthy. They employ a powerful, sophisticated approach to imaging SOS distribution within the framework of ultrasound computed tomography as well as diffraction tomography. Their experimental and clinical imaging results are promising. By surrounding a breast with 256 transducers positioned on a ring in a clinical prototype system called computed ultrasound risk evaluation (CURE), they imaged three parameters of the breast, i.e., SOS, attenuation, and reflectivity, and fused them together to present as a single pseudocolor image. The resolution was approximately 3 mm corresponding to three wavelengths at a central transducer frequency of 1.5 MHz. If we use two array transducers in transmission mode, SOS and attenuation images can be reconstructed by measuring the arrival time and amplitude of the received signal on one array transducer with insonification from the other array transducer. Figure 2 shows an imaging configuration for two opposing linear array transducers. By exchanging the transmitting and receiving roles of the two array transducers, the range of angles over which data can be acquired is doubled compared with the case of using a single array transducer and a plane reflector (Huang et al. 2004; Krueger et al. 1996). To reconstruct SOS images from the arrival time of received signals, the backprojection algorithm is employed (Glover et al. 1977; Mizutani et al. 1997). We will briefly explain the procedure used here. The time it takes for the transmit ultrasound signal from the nth
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
639
STRAIN IMAGING
Fig. 2. Transmission and reception of ultrasound signal in transmission mode.
transmit element to arrive at the receive element after traveling a distance of Lq at an angle of q with respect to the transmit array transducer normal is expressed as
e
dl ; tðq; nÞ5 Lq vðx; yÞ
(1)
where vðx; yÞ is the SOS at an imaging point ðx; yÞ: Letting vo denote the SOS in the background medium, the relative TOF can be written as
e
e
dl dl 2 : Pðq; nÞ5tðq; nÞ2to ðq; nÞ5 Lq vðx; yÞ Lq vo
(2)
Letting a distribution, f ðx; yÞ, related to the SOS, be equal to f ðx; yÞ5
1 1 2 ; vðx; yÞ vo
(3)
we can represent it in terms of the projection data as follows: Pðq; nÞ5
e f ðx; yÞdl:
(4)
Elastography is useful in imaging the mechanical properties of tumor or cancer in soft tissue such as the breast. To obtain a strain image, mechanical compression needs to be applied to the human tissue. Stain image formation proceeds by first estimating tissue displacements due to applied compression and then taking the difference of displacements along the axial or lateral direction (Ophir et al. 1991). Thus, displacement estimation is a crucial step in strain imaging. In this article, complex baseband signals are used to estimate the displacement between pre- and postcompression signals reflected from tissue (Yoon et al. 2006). The displacement is modeled using an allpass filter whose phase varies linearly with frequency. In this case, the linear phase delay is the same as the group delay. Also, both are identical to the time delay, t. If an input, x1 ðtÞ5rðtÞcosðu0 t1fðtÞÞ, is applied to the filter, then the output, which is a delayed version of the input, is given by x2 ðtÞ5x1 ðt2tÞ5rðt2tÞcosðu0 ðt2tÞ1fðt2tÞÞ;
(6)
where rðtÞ is the envelope, u0 is the radian center frequency, and fðtÞ is the time-varying phase. Following demodulation, their complex baseband signals can be expressed as xb1 ðtÞ5rðtÞejfðtÞ : xb2 ðtÞ5rðt2tÞejð2u0 t1fðt2tÞÞ
(7)
The phase difference between the two signals in a finite interval can be written as DF0 5arg,xb1 ðnÞ,xb2 ðnÞ.5u0 t1fðtÞ2fðt2tÞ; (8) where arg denotes the phase of the argument, ,,., the inner product, and , the complex conjugate. Expanding fðt2tÞ in a first-order Taylor series yields fðt2tÞzfðtÞ2tf0 ðtÞ:
(9)
Thus, eqn (8) can be simplified as
Lq
Then the SOS distribution, vðx; yÞ, can be obtained from eqn (3) as follows: 1 ; vðx; yÞ5 af ðx; yÞ11=vo
(5)
where a is a scale factor. Due to the limited view angle available for our configuration, it turns out that the SOS tends to be somewhat underestimated. Thus, to remedy this underestimation problem, the scale factor is incorporated into the above expression. It is determined experimentally.
DF0 5u0 t1f0 ðtÞt:
(10)
Solving the above equation for t, we obtain t5
DF0 DF0 5 ; u0 1f0 ðtÞ u0 1uB ðtÞ
(11)
where f0 ðtÞ5uB ðtÞ is the derivative of phase with respect to time, corresponding to the instantaneous frequency. Hence, f0 ðtÞ5uB ðtÞ5
arg,xb1 ðnÞ,xb1 ðn1TÞ. ; T
(12)
640
Ultrasound in Medicine and Biology
where T is the sampling period. This term is responsible for reducing errors associated with the center frequency shift with increasing imaging depth due to speckle and attenuation characteristics. If the displacement between the two signals to be compared is large, their decorrelation becomes large, resulting in large errors in estimating the phase difference. To make the phase difference as small as possible, our method of displacement estimation proceeds by shifting the postcompression signal, xb2 ðtÞ, by an amount corresponding to a previously estimated coarse displacement, d, as in the following expression: xb2 ðt1dÞ5rðt2t1dÞejð2u0 t1fðt2t1dÞÞ :
(14)
Consequently, we obtain DFdd t2d5 ; u0 1f0 ðtÞ
a b1
b b2
(13)
The above process is graphically depicted in Figure 3. Note that the waveforms in Figure 3 represent either the in-phase or quadrature components. Considering the fact that for proper operation, jt2dj should be maintained less than one half of the period of the signal of interest, this shifting operation leads us to obtain valid displacement estimates. Using the relationship that a temporal shift of d is equivalent to a phase rotation of ejð2u0 dÞ and expanding in a Taylor series, we can express eqn (8) as follows: DFdd 5arg ,xb1 ðtÞ,xb2 ðt1dÞ.,ejð2uo dÞ zu0 ðt2dÞ1f0 ðtÞðt2dÞ:
Volume 36, Number 4, 2010
c b2
Fig. 3. (a) Precompression signal, (b) postcompression signal, and (c) shifted version of (b) by an amount equal to the previously estimated delay.
(CNRe) values of strain image by employing adaptive strain estimation or stretching techniques (Konofagou et al. 1997; Lubinski et al. 1999; Varghese and Ophir 1997). Also, in the configuration where both array transducers face each other, the strain image can be obtained from each of both array transducers, thus making it feasible to apply the compounding operation to obtain better image quality.
(15)
where t can be correctly determined from DFdd as long as the latter is within the range ½2p; p so as to avoid aliasing. To obtain satisfactory displacement estimates, it is essential that d should be as close to t as possible. Since the displacement estimation starts at a depth of zero, as one gets deeper into the axial direction, the displacements of the current and previous data windows tend to be nearly identical. In subsequent experiments, to obtain the current displacement estimate, t, we have used the previously estimated version of displacement, t, for d. In static palpation method, the compression is externally applied manually by slowly pushing a transducer on the tissue (Hall 2004). When using two array transducers that face each other, the spacing between them can be adjusted. By changing the spacing, the amount of mechanical compression applied to the breast can be varied. This makes it possible to obtain strain images. Because the spacing between the two array transducers is finely controllable electromechanically, the amount of compression used to obtain every image frame is available, paving the way for improving the elastographic signal-tonoise ratio (SNRe) and elastographic contrast-to-noise ratio
EXPERIMENTS AND RESULTS Figure 4 shows the block diagram of an experimental setup. Figure 4a illustrates a configuration used to obtain both compound and strain images, where the two array transducers are connected to a clinical ultrasound scanner (Accuvix; Medison, Seoul, Korea) and the acquired data are transferred to a PC for signal processing. In strain imaging, the position of the lower array transducer is fixed, while the upper array transducer is movable up and down under the control of a stepper motor with a precision of 25 mm. Figure 4b shows a configuration where the two array transducers are connected to a pulser and receiver (Olympus 5077PR; Olympus NDT, Waltham, MA, USA) to obtain SOS images. The individual elements of the two array transducers are connected to a multiplexer switchboard controlled by a PC. Two 38.4 mm wide, 192 element, 7.5 MHz linear array transducers are configured to face each other with a phantom placed between them. Figure 4c is a photograph of the experimental setup used. To assess the image quality, we fabricated two types (A and B) of ultrasound phantoms in our laboratory. Phantom A for compound and SOS imaging has two
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
a Array XDCR PC
Ultrasonic imaging system
Phantom anto Array ay XDCR XD
b Array XDCR Pulser
MUX
Recei -ver
MUX
PC Oscillo -scope
Phantom Array XDCR
c
Fig. 4. Experimental setup for (a) compound and strain imaging and (b) speed of sound (SOS) imaging with (c) being the photograph.
urethane rubber cylinders with an SOS of 1450 m/s and phantom B for strain imaging contains a single cylindrical hard inclusion. The two cylinders used in making phantom A were taken out of a commercial multipurpose phantom (Model 539; ATS Laboratories, Bridgeport, CT, USA) and had diameters 7 mm and 10 mm. Both cylinders were embedded in a 40 mm thick background of plastic formed by mixing together plastic softer and hardener (M-F Manufacturing, Ft. Worth, TX, USA). The SOS in the background was 1400 m/s. The background material is mainly used to make artificial soft plastic fishing lures. Glass particles with a diameter of 27 mm (Spheriglass 2000 solid glass microspheres; Potters Industries,
641
Malvern, PA, USA) were added in the background as scatterers. The particles had no color and accounted for 0.5% of the total weight. The two parallel cylinders were positioned in perpendicular to the transducer scan plane and were displayed in ultrasound image in the form of a circle. Because the cylinders that simulate lesion do not contain scatterers, they appear darker than the background in ultrasound B-mode image, and thus it is relatively easy to identify them. To obtain compound images, ultrasound beams were steered at angles of –14 , –7 , 0 , 7 and 14 . The radiofrequency (RF) data were acquired from each of both array transducers, and were transferred to a PC, where compounding was carried out. It is to be added that actually, the steering angle of 0 corresponds to the case of no steering. The compound images were obtained by geometrically aligning five datasets, each steered at different angles, using affine transform and averaging them. The left and right panels in Figure 5 represent the B-mode and compound images, respectively. The images in the top row are obtained from the upper array transducer, those in the middle row are obtained from the lower array transducer, and the image in the bottom row shows the result of compounding the right panel images in the first and second rows. Here, the B-mode images are obtained at a steering angle of 0 , i.e., without steering. To assess the image quality between different images, the SNR was evaluated in the region marked by a rectangular box as shown in the left panel of the top row in Figure 5. It is defined as the average to the standard deviation of the pixel values in the region of interest (Wilhjelm et al. 2000). The SNRs for the top and middle row B-mode images obtained without steering were found to be 15.2 dB and 15.9 dB, respectively, and those for the compounded images in the top and middle rows were 19.0 dB and 18.9 dB, respectively. We can see that the compounding of the compound images obtained from each of both array transducers helps increase the SNR as expected because compound images obtained from each of both array transducers are essentially uncorrelated. The final compounded image is presented in the right panel of the bottom row in Figure 5 and is found to have an SNR of 21.9 dB. We also obtained SOS images for the same phantom that was used to obtain the compound images. In the SOS imaging, a multiplexer board was used between the two array transducers so that when one of the two arrays acts as a transmitter, the other can be configured to function as a receiver and vice versa. The multiplexer board can selectively connect each of the array transducer elements to the data acquisition system interfaced to a PC for data processing. The measurement was repeated for each of the 192 elements in the transmit array transducers, and after
642
Ultrasound in Medicine and Biology
Fig. 5. The images in the top and middle rows are obtained using the upper and lower linear array transducers, respectively. The left and right panels represent the B-mode and compound images, respectively. The right panel in the bottom row is the result of compounding the images in the two right panels of the top and middle rows. The two cylinders have diameters of 10 mm and 7 mm.
exchanging the role of transmission and reception, the entire process was repeated again. To reduce the data acquisition time, for every transmit element, the data were acquired from every fifth receive element starting from the center element. The interelement spacing of the transducer used is 0.2 mm. Because the data were acquired from every fifth transducer element, the SOS image was reconstructed with a resolution cell of size 1 mm 3 1 mm. To reduce noise by averaging, the data acquisition process was repeated eight times. The TOF of an ultrasound signal emitted from a selected transmit element to reach receive elements that make angles of less than 34 to the transmit array normal (see Fig. 2) was estimated using crosscorrelation on RF data acquired at a rate of 125 MHz. The acquired RF data were interpolated to a sampling rate of 1.75 GHz to improve the accuracy of TOF estimation. By estimating the absolute TOF from the signal on some transmit element and the signal on some receive element of interest and also the relative time delay between the signal on that receive element and the signal
Volume 36, Number 4, 2010
on any other receive element of interest, using crosscorrelation, we determined the TOFs from that transmit element to any other receive elements. Figure 6a shows the resulting SOS image. In reconstructing the SOS image, we used a value of 1400 m/s for vo , which is the SOS in the background, and a value of 7.3 for a, in eqn (5). Although the cylinders whose SOS is different from that of the surrounding background can be identified upon close inspection of the figure, their circular edges are not well defined. The cylinders look like ellipses because the data are acquired over a limited angle range. The two cylinders with different diameters of 7 mm and 10 mm were fabricated to have the same SOS but it was found that the smaller one had a lower SOS. In Figure 6b, the dotted lines represent the true SOS profile of the two cylinders. The solid and dashed lines depict the reconstructed SOS profiles along the lateral direction that pass through the centers of the left and right cylinders in Figure 6a, respectively. It can be seen that there is still room for improvement in the overall performance of SOS estimation due to the fact that the data are acquired only over a limited view of the object. However, the experimental results show that we are able to obtain an SOS contrast necessary to differentiate between the cylinders and the background. For strain imaging, phantom B, which is an elasticity phantom with a single inclusion, was also fabricated in our laboratory. The stiffness of the inclusion was controlled by varying the amount of the plastic hardener and softener (M-F Manufacturing, Ft. Worth, TX, USA) that were mixed together. Phantom B had a 10 mm diameter cylindrical inclusion with a Young’s modulus of 58.5 kPa embedded at a depth of 15 mm in the background with a Young’s modulus of 11.1 kPa. The Young’s moduli were measured using Erkamp et al.’s method (Erkamp et al. 1998). The total height of the elasticity phantom was 37 mm. The elasticity phantom was placed between the upper and lower array transducers. As the upper array transducer was moved downward using a stepper motor in steps of 0.42% in strain, the RF datasets before and after applying compression were acquired from each of both array transducers. The top and middle panels in Figure 7 represent strain images obtained with each of both array transducers facing each other and the resulting compounded strain image is shown in the bottom panel of Figure 7. Note that the cylindrical inclusion is of diameter 10 mm. The strain images were obtained by employing Yoon et al.’s method (Yoon et al. 2006) described in the previous section. In our present method, because the spacing between the array transducers can be finely controlled by a stepper motor, the adaptive strain estimation technique can be applied to the successive frame data obtained at different
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
0
a
643
1450 1445 1440
10
1435 1430 1425
20
1420 1415 1410
30 mm
1405 1400
0
b
10
20
30
40 mm
1460 1450
speed [m/s]
1440 1430 1420 1410 1400 1390
0
5
10
15
20
25
30
35
distance [mm]
Fig. 6. (a) The reconstructed speed of sound (SOS) image of a phantom with two embedded cylinders. The diameters of the left and right cylinders are 10 mm and 7 mm, respectively. The gray-scale map is in units of m/s. (b) The dotted lines represent the true SOS profile of the two cylinders. The solid and dashed lines depict the reconstructed SOS profiles along the lateral direction that pass through the centers of the left and right cylinders in image (a), respectively.
steps of compression (Konofagou et al. 1997; Lubinski et al. 1999). However, the effect of averaging is not significant because there is large correlation between the strain images of the same region subjected to different amounts of compression applied by only one array transducer. On the contrary, we were able to improve the strain image quality by compounding the strain images obtained in opposite directions using each of the two array transducers. The strain image quality is evaluated by the SNRe and CNRe which are defined as follows (Cespedes and Ophir 1993; Bilgen and Insana 1997; Lopata et al. 2009): SNRe 520 log10 msbb iÞ CNRe 520 log10 ðmsb22m 1s2 b
i
2
;
(16)
Fig. 7. The strain image of the cylindrical inclusion phantom when the applied strain is 0.42%. The top and middle panels show strain images obtained from the upper and the lower array transducer, respectively, with the compounded strain image shown in the bottom panel. The cylindrical inclusion is of diameter 10 mm.
where the subscripts b and i indicate the background and the inclusion, respectively, and m and s denote the mean and standard deviation of the estimated strain in a selected region of interest, respectively. Table 1 presents the SNRe and CNRe values computed to compare the quality of the compounded strain images when the applied strain is 0.42% (Techavipoo and Varghese 2005). The SNRe values were computed in region A indicated by the two black windows in Fig. 7 and the CNRe values were computed for region A relative to region B indicated by the white window in Fig. 7. We can see from Table 1 that the compound strain images has somewhat better SNRe and CNRe values than either of the individual uncompounded strain images.
644
Ultrasound in Medicine and Biology
Table 1. Comparison of the SNRe and CNRe values when the applied strain is 0.42%.
Top view Bottom view Compounding
SNRe (dB)
CNRe (dB)
25.17 23.27 27.59
51.45 47.95 56.56
SNRe 5 elastographic signal-to-noise ratio; CNRe 5 elastographic contrast-to-noise ratio.
To assess the effect of compounding on strain image quality under different amounts of compression, multiple consecutive image frames were acquired while compression was being applied mechanically. Every pair of two successive image frames was made to be different in the
a
29
28
SNRe [dB]
27
26
25
24
23 0.2
0.4
0.6
0.8
1
1.2
1.4
1
1.2
1.4
strain [% ]
b
58
56
CNRe [dB]
54
52
50
48
46 0.2
0.4
0.6
0.8 strain [% ]
Fig. 8. Comparison of the (a) elastographic signal-to-noise ratio (SNRe) and (b) elastographic contrast-to-noise ratio (CNRe) values as the amount of strain applied is increased. The solid and dotted lines represent the cases with and without compounding, respectively. The error bars denote one standard deviation.
Volume 36, Number 4, 2010
amount of compression by 0.42%. Images that are two frames apart were taken to make the amount of compression equal to 0.84% and their strain was estimated to construct strain image. Also, images that are three frames apart were taken to make the amount of compression equal to 1.26% and their strain is estimated to construct strain image. Figure 8 shows the effect of compounding as the applied strain is variably set to 0.42%, 0.84%, and 1.26%. The dotted lines represent the case of using either of the two array transducers while the solid line represents the case of compounding both array transducers. Figures 8a and b plot the SNRe and CNRe values, respectively, as the applied strain is increased. The vertical error bars denote one standard deviation from the mean. The SNRe value of the compounded strain image has increased by more than 2.5 dB compared with the uncompounded strain images. It is known that in general, increasing the amount of applied strain to within some range tends to improve the SNRe value. It is found, however, that compounding of strain images is more effective in reducing noise (Lubinski et al. 1999). The strain image quality can also be improved by averaging over consecutive image frames based on persistence. Persistence is a signal processing method widely used to improve ultrasound image quality by averaging consecutive moving image frames and thereby reducing noise (Wang and Liu 2006). In the method, the noise can be suppressed considerably but the spatial resolution is degraded because the constituent strain image frames do not exactly match geometrically. Those image frames need to be geometrically matched using an affine transform. In general, in a manual palpation method, the amount of strain applied cannot be exactly estimated so the geometrical distortion cannot be compensated for exactly. On the contrary, in our proposed method, the amount of strain applied is exactly known and, thus, the distortion can be compensated for. Figure 9 plots the strain profiles along the scan lines that pass through the center of the cylinder in frames 1, 6, and 11 of strain images which are obtained successively, as a constant strain of 0.42% is applied from one frame to the next in all steps of the data acquisition process. The strain between frames 1 and 2 is 0.42%, that between frames 1 and 3 is 0.84%, that between frames 1 and 4 is 1.26%, and so on. The strain increases with increasing frame number. The strain values inside the cylinder are shown to be small. However, neither the falling nor the rising edges shown in Figure 9a coincide exactly. Figure 9b shows the result of performing a global stretching operation in Figure 9a using the applied strain where both left and right edges are seen to be aligned. Averaging these aligned strain images results in a decrease of noise without degrading the image resolution.
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
645
1.8 1.6
relative magnitude
1.4 1.2 1 0.8 0.6 0.4 0.2 0
0
50
100
150
200
250
pixel
Fig. 10. Strain image after global stretching is applied to RF data. The cylindrical inclusion is of diameter 10 mm.
1.8 1.6
relative magnitude
1.4 1.2 1 0.8 0.6 0.4 0.2 0
0
50
100
150
200
250
pixel
Fig. 9. Strain profiles along the scan lines that pass through the center of the cylinder in frames 1, 6 and 11 of consecutively obtained strain images by increasing the applied strain in steps of 0.42% with increasing frame number without (top) and with (bottom) global stretching.
suitable for imaging protruding objects such as a human breast placed between two opposing linear array transducers. In vitro images of compounding and SOS as well as strain are reconstructed. Although there is some distortion in the SOS images due to limited view angle, those three types of images can be used complementarily to aid in better diagnosis. Unlike computed or diffraction tomography that uses water to surround a breast, our imaging method is advantageous in that there is no need to correct for refraction at the surface of it since the transducers are in direct contact with it. The surface refraction is one of the major sources of error in SOS estimation. The data required for the proposed method can be acquired very fast if a dedicated hardware system that operates in real time is built. Therefore, the diagnosis can be made within a small amount of time without causing much discomfort to the patient. Acknowledgements—This work was supported by Medison.
Figure 10 shows the result of applying global stretching to the RF data where the applied strain is 0.42%. The diameter of the cylindrical inclusion is 10 mm. The soft region is stretched to equal its original length, but the hard region is expanded because the latter has not been compressed much. Thus, the black and white levels are reversed so that the hard region appears to be brighter. This characteristic helps us better identify the hard region. Our method has the advantage that the amount of the total compression is available. CONCLUSIONS We have investigated a new linear array transducer configuration for compound, SOS, and strain imaging. Its feasibility was confirmed by experiments with two types of in-house phantoms. The proposed method is
REFERENCES Bilgen M, Insana MF. Predicting target detectability in acoustic elastography. Proc IEEE Ultrason Symp 1997;1427–1430. Carson PL, Meyer CR, Scherzinger AL, Oughton TV. Breast imaging in coronal planes with simultaneous pulse echo and transmission ultrasound. Science 1981;214:1141–1143. Cespedes I, Ophir J. Reduction of image noise in elastography. Ultrason Imag 1993;15:89–102. Duric N, Li C, Glide-Hurst C, Littrup P, Huang L, Lupinacci J, Schmidt S, Rama O, Bey-Knight L, Xu Y. Breast imaging with ultrasound tomography: Clinical results at the Karmanos Cancer Institute. 2008 Int Conf Biomed Eng Informatics 2008;713–717. Erkamp RQ, Wiggins P, Skovoroda AR, Emelianov SY, O’Donnell M. Measuring the elastic modulus of small tissue samples. Ultrason Imag 1998;20:17–28. Glover GH, Sharp JC. Reconstruction of ultrasound propagation speed distributions in soft tissue: Time-of-flight tomography. IEEE Trans Sonic Ultrason 1977;24:229–234. Goss SA, Johnston RL, Dunn F. Comprehensive compilation of empirical ultrasonic properties of mammalian tissues. J Acoust Soc Amer 1978;64:423–457.
646
Ultrasound in Medicine and Biology
Greenleaf JF, Johnson SA, Samoyoa WF, Duck FA. Algebraic reconstruction of spatial distributions of acoustic velocities in tissue from their time-of-flight profiles. Acoust Holography 1975;6:71–90. Greenleaf JF, Bahn RC. Clinical imaging with transmissive ultrasonic computerized tomography. IEEE Trans Biomed Eng 1981;28:177–185. Hall TJ. The physics of elasticity imaging: A new option on the latest US systems. Proc RSNA 2004;233–246. Hansen C, Schasse A, Hu¨tterbra¨uker N, Ashfaq M, Wilkening W, Ermert H. Reconstruction of speed of sound for a correction of transit time in full angle spatial compounding. Proc IEEE Ultrason Symp 2007;785–788. Huang S-W, Li P-C. Computed tomography sound velocity reconstruction using incomplete data. IEEE Trans Ultrason Ferroelectr Freq Control 2004;51. 329–243. Jeong M-K, Kwon S-J, Cho S-M, Bae M-H, Kim Y-G. Ultrasound breast imaging technique using two opposing transducer arrays. Proc IEEE Ultrason Symp 2008;1274–1277. Konofagou EE, Ophir J, Kaller F, Varghese T. Elastographic dynamic range expansion using variable applied strains. Ultrason Imag 1997;19:145–166. Krueger M, Pesavento A, Ermert H. A modified time-of-flight tomography concept for ultrasonic breast imaging. Proc IEEE Ultrason Symp 1996;1381–1385. Kru¨cker JF, Fowlkes JB, Carson PL. Sound speed estimation using automatic ultrasound image registration. IEEE Trans Ultrason Ferroelectr Freq Control 2004;51:1095–1106. Li C, Duric N, Huang L. Breast imaging using transmission ultrasound: Reconstructing tissue parameters of sound speed and attenuation. 2008 Int Conf Biomed Eng Informatics 2008;708–712. Li C, Duric N, Littrup P, Huang L. In vivo breast sound-speed imaging with ultrasound tomography. Ultrasound Med Biol 2009;35: 1615–1628. Lopata RGP, Nillesen MM, Hendrik HGH, Gerrits IH, Thijssen JM, de Korte CL. Performance evaluation of methods for two-dimensional
Volume 36, Number 4, 2010 displacement and strain estimation using ultrasound radio frequency data. Ultrasound Med Biol 2009;35:796–812. Lubinski MA, Emelianov SY, O’Donnell M. Adaptive strain estimation using retrospective processing. IEEE Trans Ultrason Ferroelectr Freq Control 1999;46:97–107. Mizutani K, Nishizaki K, Nagai K, Harakawa K. Measurement of temperature distribution in space using ultrasound computerized tomography. Jpn J Appl Phys 1997;36:3176–3177. Ophir J, Ce´spedes I, Ponnekanti H, Yazdi Y, Li X. Elastography: A quantitative method for imaging the elasticity of biological tissues. Ultrason Imag 1991;13:111–134. Robinson DE. Digital reconstruction and display of compound scan ultrasound images. IEEE Trans Sonics Ultrason 1984;31: 396–406. Robinson DE, Ophir J, Wilson LS, Chen CF. Pulse-echo ultrasound speed measurements: Progress and prospects. Ultrasound Med Biol 1991;17:633–646. Sarvazyan AP, Rudenko OV, Swanson SD, Fowlkes JB, Emelianov SY. Shear wave elasticity imaging: A new ultrasonic technology of medical diagnostics. Ultrasound Med Biol 1998;24:1419–1435. Techavipoo U, Varghese T. Improvements in elastographic contrast-tonoise ratio using spatial-angular compounding. Ultrasound Med Biol 2005;31:529–536. Varghese T, Ophir J. Enhancement of echo-signal correlation in elastography using temporal stretching. IEEE Trans Ultrason Ferroelectr Freq Control 1997;44:173–180. Wang G, Liu DC. Adaptive persistence utilizing motion compensation for ultrasound images. Proc 18th Int Conf Pattern Recognition 2006;3:726–729. Wilhjelm JE, Jensen MS, Brandt T, Sahl B, Martinsen K, Jespersen SK, Falk E. Some imaging strategies in multi-angle spatial compounding. Proc IEEE Ultrason Symp 2000;1615–1618. Yoon R-Y, Kwon S-J, Bae M-H, Jeong M-K. Improved ultrasonic elasticity imaging with center frequency estimation and global shift compensation. Proc IEEE Ultrason Symp 2006;1278–1281.
doi:10.1016/j.ultrasmedbio.2010.01.006
d
Original Contribution MULTIMODE ULTRASOUND BREAST IMAGING USING A NEW ARRAY TRANSDUCER CONFIGURATION* MOK-KUN JEONG and SUNG-JAE KWON Departments of Electronic and Communication Engineering, Daejin University, Pocheon, Gyeonggi, Korea (Received 1 June 2009; revised 22 December 2009; in final form 22 January 2010)
Abstract—This article presents a diagnostic ultrasound imaging technique that can be used in imaging protruding objects such as a human breast using two opposing array transducers. Because two B-mode images obtained from each of the two linear array transducers facing each other represent the same imaging area viewed in different directions, the image quality can be improved using a compounding technique. Using one array as a transmitter and the other as a receiver, the speed of sound distribution in a medium interposed between them is also reconstructed. In addition, because the spacing between the two arrays can be finely controlled, strain image can also be obtained. This new method can be used to produce a compound B-mode image, a speed of sound image, and a strain image of the same region-of-interest, making it possible to obtain more information leading to better diagnosis. Experimental results on a phantom containing a cylinder of different speed of sound and elasticity confirm that the proposed method is useful in obtaining compound and speed of sound images as well as strain images. (E-mail: [email protected]) Ó 2010 World Federation for Ultrasound in Medicine & Biology. Key Words: Attenuation, Backprojection, Breast, Compounding, Reconstruction, Speckle, Speed of sound, Strain imaging.
relatively long data acquisition time and discomfort to patients, it has not yet been put to practical use. To overcome these limitations, several researchers proposed data acquisition methods based on linear array transducers (Huang et al. 2004; Krueger et al. 1996). As with mammography, using a linear array transducer and a plane reflector, the SOS image of a human tissue positioned between them was obtained. Compared with tomographic reconstruction methods that collect data over an angle of 360 , the linear array method acquires data over a limited range of angles, making it difficult to produce a complete artifact-free image (Krueger et al. 1996). Huang et al. (2004) compensated for distortion in SOS image using structural information extracted from B-mode image. To superpose images obtained from multiple viewing angles in full angle spatial compounding, Hansen et al. (2007) reconstructed the SOS distribution by placing a reflector behind an object of interest and using the filtered backprojection of the echo data from the reflector. Jeong et al. (2008) obtained SOS images in transmission mode using two transducers that face each other. Compound imaging techniques acquire data using spatial or frequency diversity and average images of the same region to achieve less speckle noise and better contrast
INTRODUCTION Although mammography is widely used in the diagnosis of breast cancer, it may be subject to potential radiation hazard. Breast cancer has different ultrasound characteristics from normal healthy tissue in terms of speed of sound (SOS) and attenuation (Goss et al. 1978). For the case of diagnosing breasts in medical ultrasound imaging, tissue parameter imaging can be used together with B-mode imaging in a complementary manner. Since the breast protrudes from a human body, it is possible to obtain attenuation and SOS images by using circular transducers or by rotating linear arrays over 360 in transmission or reflection mode using tomographic reconstruction principles (Glover et al. 1977; Greenleaf et al. 1975, 1981). Although this approach enables us to obtain more information about lesion, the use of circular transducers or linear arrays entails the immersion of objects to be imaged in water. Due to
Address correspondence to: Prof. Sung-Jae Kwon, Departments of Electronic and Communication Engineering, Daejin University, Pocheon, Gyeonggi 487-711, Korea. E-mail: [email protected] * Part of this paper was presented at 2008 IEEE Ultrasonics Symposium, Beijing, China, by Jeong et al. under the title ‘‘Ultrasound breast imaging technique using two opposing transducer arrays.’’ 637
638
Ultrasound in Medicine and Biology
(Robinson 1984; Wilhjelm et al. 2000). Carson et al. (1981) introduced a method of obtaining compound image together with SOS image using circular array transducers. Recently, there has been much research into elasticity imaging of the breast. Since tumor or cancer in soft tissue such as the breast and prostate tends to be stiffer than the surrounding tissue, elasticity imaging that visualizes the degree of stiffness is of great help in diagnosing it (Hall 2004; Ophir et al. 1991; Sarvazyan et al. 1998). Human tissues tend to exhibit a higher contrast in elasticity than other tissue parameters. Therefore, the elasticity parameter is amenable to imaging. In this article, we obtain spatial compound, SOS, and strain images using two opposing linear array transducers. B-mode or strain images are obtained independently in reflection mode from each of the two array transducers, and are compounded to improve the image quality. In transmission mode, however, we use an imaging configuration where one array transducer is used as a transmitter and the other is used as a receiver. The time of flight (TOF) is measured and an SOS image is reconstructed. A combination of the B-mode, strain, and SOS imaging can increase the accuracy of lesion diagnosis. This article presents each imaging method and experimental results. SPATIAL COMPOUND IMAGING Compound imaging is used to reduce speckle noise as well as artifacts such as shadowing and reverberation by averaging a certain number of images whose correlations are small. However, the degradation of resolution and signal-to-noise ratio (SNR) with increasing depth cannot be compensated for. Conventional spatial compound B-mode images are produced by averaging individual images of the same imaging region obtained by steering scan lines at different angles. However, the number of useful images obtainable with a single transducer array is limited by its characteristics or the performance of an ultrasound imaging system employing it. If two linear array transducers are arranged such that they face each other, compound images can be reconstructed separately using each of them and those compound images can be averaged again. As a result, the images can be compensated for their depth-dependent characteristics. Figure 1 shows the imaging configuration used in this article where the two linear array transducers face each other. A slanted imaging area that makes an angle of q to the upper transducer array normal by steering ultrasound beams is also indicated in dashed lines in the figure. SOS IMAGING The measurement of ultrasound speed has been a topic of interest since it can aid in tissue characterization, beamforming, scan conversion, and image registration
Volume 36, Number 4, 2010
Fig. 1. Compounding of independently obtained images using each of two opposing array transducers.
(Duric et al. 2008; Glover et al. 1977; Goss et al. 1978; Greenleaf et al. 1975, 1981; Hansen et al. 2007; Huang et al. 2004; Krueger et al. 1996; Kru¨cker et al. 2004; Li 2008, 2009; Robinson et al. 1991). Recently, the work of Duric et al. has been quite noteworthy. They employ a powerful, sophisticated approach to imaging SOS distribution within the framework of ultrasound computed tomography as well as diffraction tomography. Their experimental and clinical imaging results are promising. By surrounding a breast with 256 transducers positioned on a ring in a clinical prototype system called computed ultrasound risk evaluation (CURE), they imaged three parameters of the breast, i.e., SOS, attenuation, and reflectivity, and fused them together to present as a single pseudocolor image. The resolution was approximately 3 mm corresponding to three wavelengths at a central transducer frequency of 1.5 MHz. If we use two array transducers in transmission mode, SOS and attenuation images can be reconstructed by measuring the arrival time and amplitude of the received signal on one array transducer with insonification from the other array transducer. Figure 2 shows an imaging configuration for two opposing linear array transducers. By exchanging the transmitting and receiving roles of the two array transducers, the range of angles over which data can be acquired is doubled compared with the case of using a single array transducer and a plane reflector (Huang et al. 2004; Krueger et al. 1996). To reconstruct SOS images from the arrival time of received signals, the backprojection algorithm is employed (Glover et al. 1977; Mizutani et al. 1997). We will briefly explain the procedure used here. The time it takes for the transmit ultrasound signal from the nth
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
639
STRAIN IMAGING
Fig. 2. Transmission and reception of ultrasound signal in transmission mode.
transmit element to arrive at the receive element after traveling a distance of Lq at an angle of q with respect to the transmit array transducer normal is expressed as
e
dl ; tðq; nÞ5 Lq vðx; yÞ
(1)
where vðx; yÞ is the SOS at an imaging point ðx; yÞ: Letting vo denote the SOS in the background medium, the relative TOF can be written as
e
e
dl dl 2 : Pðq; nÞ5tðq; nÞ2to ðq; nÞ5 Lq vðx; yÞ Lq vo
(2)
Letting a distribution, f ðx; yÞ, related to the SOS, be equal to f ðx; yÞ5
1 1 2 ; vðx; yÞ vo
(3)
we can represent it in terms of the projection data as follows: Pðq; nÞ5
e f ðx; yÞdl:
(4)
Elastography is useful in imaging the mechanical properties of tumor or cancer in soft tissue such as the breast. To obtain a strain image, mechanical compression needs to be applied to the human tissue. Stain image formation proceeds by first estimating tissue displacements due to applied compression and then taking the difference of displacements along the axial or lateral direction (Ophir et al. 1991). Thus, displacement estimation is a crucial step in strain imaging. In this article, complex baseband signals are used to estimate the displacement between pre- and postcompression signals reflected from tissue (Yoon et al. 2006). The displacement is modeled using an allpass filter whose phase varies linearly with frequency. In this case, the linear phase delay is the same as the group delay. Also, both are identical to the time delay, t. If an input, x1 ðtÞ5rðtÞcosðu0 t1fðtÞÞ, is applied to the filter, then the output, which is a delayed version of the input, is given by x2 ðtÞ5x1 ðt2tÞ5rðt2tÞcosðu0 ðt2tÞ1fðt2tÞÞ;
(6)
where rðtÞ is the envelope, u0 is the radian center frequency, and fðtÞ is the time-varying phase. Following demodulation, their complex baseband signals can be expressed as xb1 ðtÞ5rðtÞejfðtÞ : xb2 ðtÞ5rðt2tÞejð2u0 t1fðt2tÞÞ
(7)
The phase difference between the two signals in a finite interval can be written as DF0 5arg,xb1 ðnÞ,xb2 ðnÞ.5u0 t1fðtÞ2fðt2tÞ; (8) where arg denotes the phase of the argument, ,,., the inner product, and , the complex conjugate. Expanding fðt2tÞ in a first-order Taylor series yields fðt2tÞzfðtÞ2tf0 ðtÞ:
(9)
Thus, eqn (8) can be simplified as
Lq
Then the SOS distribution, vðx; yÞ, can be obtained from eqn (3) as follows: 1 ; vðx; yÞ5 af ðx; yÞ11=vo
(5)
where a is a scale factor. Due to the limited view angle available for our configuration, it turns out that the SOS tends to be somewhat underestimated. Thus, to remedy this underestimation problem, the scale factor is incorporated into the above expression. It is determined experimentally.
DF0 5u0 t1f0 ðtÞt:
(10)
Solving the above equation for t, we obtain t5
DF0 DF0 5 ; u0 1f0 ðtÞ u0 1uB ðtÞ
(11)
where f0 ðtÞ5uB ðtÞ is the derivative of phase with respect to time, corresponding to the instantaneous frequency. Hence, f0 ðtÞ5uB ðtÞ5
arg,xb1 ðnÞ,xb1 ðn1TÞ. ; T
(12)
640
Ultrasound in Medicine and Biology
where T is the sampling period. This term is responsible for reducing errors associated with the center frequency shift with increasing imaging depth due to speckle and attenuation characteristics. If the displacement between the two signals to be compared is large, their decorrelation becomes large, resulting in large errors in estimating the phase difference. To make the phase difference as small as possible, our method of displacement estimation proceeds by shifting the postcompression signal, xb2 ðtÞ, by an amount corresponding to a previously estimated coarse displacement, d, as in the following expression: xb2 ðt1dÞ5rðt2t1dÞejð2u0 t1fðt2t1dÞÞ :
(14)
Consequently, we obtain DFdd t2d5 ; u0 1f0 ðtÞ
a b1
b b2
(13)
The above process is graphically depicted in Figure 3. Note that the waveforms in Figure 3 represent either the in-phase or quadrature components. Considering the fact that for proper operation, jt2dj should be maintained less than one half of the period of the signal of interest, this shifting operation leads us to obtain valid displacement estimates. Using the relationship that a temporal shift of d is equivalent to a phase rotation of ejð2u0 dÞ and expanding in a Taylor series, we can express eqn (8) as follows: DFdd 5arg ,xb1 ðtÞ,xb2 ðt1dÞ.,ejð2uo dÞ zu0 ðt2dÞ1f0 ðtÞðt2dÞ:
Volume 36, Number 4, 2010
c b2
Fig. 3. (a) Precompression signal, (b) postcompression signal, and (c) shifted version of (b) by an amount equal to the previously estimated delay.
(CNRe) values of strain image by employing adaptive strain estimation or stretching techniques (Konofagou et al. 1997; Lubinski et al. 1999; Varghese and Ophir 1997). Also, in the configuration where both array transducers face each other, the strain image can be obtained from each of both array transducers, thus making it feasible to apply the compounding operation to obtain better image quality.
(15)
where t can be correctly determined from DFdd as long as the latter is within the range ½2p; p so as to avoid aliasing. To obtain satisfactory displacement estimates, it is essential that d should be as close to t as possible. Since the displacement estimation starts at a depth of zero, as one gets deeper into the axial direction, the displacements of the current and previous data windows tend to be nearly identical. In subsequent experiments, to obtain the current displacement estimate, t, we have used the previously estimated version of displacement, t, for d. In static palpation method, the compression is externally applied manually by slowly pushing a transducer on the tissue (Hall 2004). When using two array transducers that face each other, the spacing between them can be adjusted. By changing the spacing, the amount of mechanical compression applied to the breast can be varied. This makes it possible to obtain strain images. Because the spacing between the two array transducers is finely controllable electromechanically, the amount of compression used to obtain every image frame is available, paving the way for improving the elastographic signal-tonoise ratio (SNRe) and elastographic contrast-to-noise ratio
EXPERIMENTS AND RESULTS Figure 4 shows the block diagram of an experimental setup. Figure 4a illustrates a configuration used to obtain both compound and strain images, where the two array transducers are connected to a clinical ultrasound scanner (Accuvix; Medison, Seoul, Korea) and the acquired data are transferred to a PC for signal processing. In strain imaging, the position of the lower array transducer is fixed, while the upper array transducer is movable up and down under the control of a stepper motor with a precision of 25 mm. Figure 4b shows a configuration where the two array transducers are connected to a pulser and receiver (Olympus 5077PR; Olympus NDT, Waltham, MA, USA) to obtain SOS images. The individual elements of the two array transducers are connected to a multiplexer switchboard controlled by a PC. Two 38.4 mm wide, 192 element, 7.5 MHz linear array transducers are configured to face each other with a phantom placed between them. Figure 4c is a photograph of the experimental setup used. To assess the image quality, we fabricated two types (A and B) of ultrasound phantoms in our laboratory. Phantom A for compound and SOS imaging has two
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
a Array XDCR PC
Ultrasonic imaging system
Phantom anto Array ay XDCR XD
b Array XDCR Pulser
MUX
Recei -ver
MUX
PC Oscillo -scope
Phantom Array XDCR
c
Fig. 4. Experimental setup for (a) compound and strain imaging and (b) speed of sound (SOS) imaging with (c) being the photograph.
urethane rubber cylinders with an SOS of 1450 m/s and phantom B for strain imaging contains a single cylindrical hard inclusion. The two cylinders used in making phantom A were taken out of a commercial multipurpose phantom (Model 539; ATS Laboratories, Bridgeport, CT, USA) and had diameters 7 mm and 10 mm. Both cylinders were embedded in a 40 mm thick background of plastic formed by mixing together plastic softer and hardener (M-F Manufacturing, Ft. Worth, TX, USA). The SOS in the background was 1400 m/s. The background material is mainly used to make artificial soft plastic fishing lures. Glass particles with a diameter of 27 mm (Spheriglass 2000 solid glass microspheres; Potters Industries,
641
Malvern, PA, USA) were added in the background as scatterers. The particles had no color and accounted for 0.5% of the total weight. The two parallel cylinders were positioned in perpendicular to the transducer scan plane and were displayed in ultrasound image in the form of a circle. Because the cylinders that simulate lesion do not contain scatterers, they appear darker than the background in ultrasound B-mode image, and thus it is relatively easy to identify them. To obtain compound images, ultrasound beams were steered at angles of –14 , –7 , 0 , 7 and 14 . The radiofrequency (RF) data were acquired from each of both array transducers, and were transferred to a PC, where compounding was carried out. It is to be added that actually, the steering angle of 0 corresponds to the case of no steering. The compound images were obtained by geometrically aligning five datasets, each steered at different angles, using affine transform and averaging them. The left and right panels in Figure 5 represent the B-mode and compound images, respectively. The images in the top row are obtained from the upper array transducer, those in the middle row are obtained from the lower array transducer, and the image in the bottom row shows the result of compounding the right panel images in the first and second rows. Here, the B-mode images are obtained at a steering angle of 0 , i.e., without steering. To assess the image quality between different images, the SNR was evaluated in the region marked by a rectangular box as shown in the left panel of the top row in Figure 5. It is defined as the average to the standard deviation of the pixel values in the region of interest (Wilhjelm et al. 2000). The SNRs for the top and middle row B-mode images obtained without steering were found to be 15.2 dB and 15.9 dB, respectively, and those for the compounded images in the top and middle rows were 19.0 dB and 18.9 dB, respectively. We can see that the compounding of the compound images obtained from each of both array transducers helps increase the SNR as expected because compound images obtained from each of both array transducers are essentially uncorrelated. The final compounded image is presented in the right panel of the bottom row in Figure 5 and is found to have an SNR of 21.9 dB. We also obtained SOS images for the same phantom that was used to obtain the compound images. In the SOS imaging, a multiplexer board was used between the two array transducers so that when one of the two arrays acts as a transmitter, the other can be configured to function as a receiver and vice versa. The multiplexer board can selectively connect each of the array transducer elements to the data acquisition system interfaced to a PC for data processing. The measurement was repeated for each of the 192 elements in the transmit array transducers, and after
642
Ultrasound in Medicine and Biology
Fig. 5. The images in the top and middle rows are obtained using the upper and lower linear array transducers, respectively. The left and right panels represent the B-mode and compound images, respectively. The right panel in the bottom row is the result of compounding the images in the two right panels of the top and middle rows. The two cylinders have diameters of 10 mm and 7 mm.
exchanging the role of transmission and reception, the entire process was repeated again. To reduce the data acquisition time, for every transmit element, the data were acquired from every fifth receive element starting from the center element. The interelement spacing of the transducer used is 0.2 mm. Because the data were acquired from every fifth transducer element, the SOS image was reconstructed with a resolution cell of size 1 mm 3 1 mm. To reduce noise by averaging, the data acquisition process was repeated eight times. The TOF of an ultrasound signal emitted from a selected transmit element to reach receive elements that make angles of less than 34 to the transmit array normal (see Fig. 2) was estimated using crosscorrelation on RF data acquired at a rate of 125 MHz. The acquired RF data were interpolated to a sampling rate of 1.75 GHz to improve the accuracy of TOF estimation. By estimating the absolute TOF from the signal on some transmit element and the signal on some receive element of interest and also the relative time delay between the signal on that receive element and the signal
Volume 36, Number 4, 2010
on any other receive element of interest, using crosscorrelation, we determined the TOFs from that transmit element to any other receive elements. Figure 6a shows the resulting SOS image. In reconstructing the SOS image, we used a value of 1400 m/s for vo , which is the SOS in the background, and a value of 7.3 for a, in eqn (5). Although the cylinders whose SOS is different from that of the surrounding background can be identified upon close inspection of the figure, their circular edges are not well defined. The cylinders look like ellipses because the data are acquired over a limited angle range. The two cylinders with different diameters of 7 mm and 10 mm were fabricated to have the same SOS but it was found that the smaller one had a lower SOS. In Figure 6b, the dotted lines represent the true SOS profile of the two cylinders. The solid and dashed lines depict the reconstructed SOS profiles along the lateral direction that pass through the centers of the left and right cylinders in Figure 6a, respectively. It can be seen that there is still room for improvement in the overall performance of SOS estimation due to the fact that the data are acquired only over a limited view of the object. However, the experimental results show that we are able to obtain an SOS contrast necessary to differentiate between the cylinders and the background. For strain imaging, phantom B, which is an elasticity phantom with a single inclusion, was also fabricated in our laboratory. The stiffness of the inclusion was controlled by varying the amount of the plastic hardener and softener (M-F Manufacturing, Ft. Worth, TX, USA) that were mixed together. Phantom B had a 10 mm diameter cylindrical inclusion with a Young’s modulus of 58.5 kPa embedded at a depth of 15 mm in the background with a Young’s modulus of 11.1 kPa. The Young’s moduli were measured using Erkamp et al.’s method (Erkamp et al. 1998). The total height of the elasticity phantom was 37 mm. The elasticity phantom was placed between the upper and lower array transducers. As the upper array transducer was moved downward using a stepper motor in steps of 0.42% in strain, the RF datasets before and after applying compression were acquired from each of both array transducers. The top and middle panels in Figure 7 represent strain images obtained with each of both array transducers facing each other and the resulting compounded strain image is shown in the bottom panel of Figure 7. Note that the cylindrical inclusion is of diameter 10 mm. The strain images were obtained by employing Yoon et al.’s method (Yoon et al. 2006) described in the previous section. In our present method, because the spacing between the array transducers can be finely controlled by a stepper motor, the adaptive strain estimation technique can be applied to the successive frame data obtained at different
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
0
a
643
1450 1445 1440
10
1435 1430 1425
20
1420 1415 1410
30 mm
1405 1400
0
b
10
20
30
40 mm
1460 1450
speed [m/s]
1440 1430 1420 1410 1400 1390
0
5
10
15
20
25
30
35
distance [mm]
Fig. 6. (a) The reconstructed speed of sound (SOS) image of a phantom with two embedded cylinders. The diameters of the left and right cylinders are 10 mm and 7 mm, respectively. The gray-scale map is in units of m/s. (b) The dotted lines represent the true SOS profile of the two cylinders. The solid and dashed lines depict the reconstructed SOS profiles along the lateral direction that pass through the centers of the left and right cylinders in image (a), respectively.
steps of compression (Konofagou et al. 1997; Lubinski et al. 1999). However, the effect of averaging is not significant because there is large correlation between the strain images of the same region subjected to different amounts of compression applied by only one array transducer. On the contrary, we were able to improve the strain image quality by compounding the strain images obtained in opposite directions using each of the two array transducers. The strain image quality is evaluated by the SNRe and CNRe which are defined as follows (Cespedes and Ophir 1993; Bilgen and Insana 1997; Lopata et al. 2009): SNRe 520 log10 msbb iÞ CNRe 520 log10 ðmsb22m 1s2 b
i
2
;
(16)
Fig. 7. The strain image of the cylindrical inclusion phantom when the applied strain is 0.42%. The top and middle panels show strain images obtained from the upper and the lower array transducer, respectively, with the compounded strain image shown in the bottom panel. The cylindrical inclusion is of diameter 10 mm.
where the subscripts b and i indicate the background and the inclusion, respectively, and m and s denote the mean and standard deviation of the estimated strain in a selected region of interest, respectively. Table 1 presents the SNRe and CNRe values computed to compare the quality of the compounded strain images when the applied strain is 0.42% (Techavipoo and Varghese 2005). The SNRe values were computed in region A indicated by the two black windows in Fig. 7 and the CNRe values were computed for region A relative to region B indicated by the white window in Fig. 7. We can see from Table 1 that the compound strain images has somewhat better SNRe and CNRe values than either of the individual uncompounded strain images.
644
Ultrasound in Medicine and Biology
Table 1. Comparison of the SNRe and CNRe values when the applied strain is 0.42%.
Top view Bottom view Compounding
SNRe (dB)
CNRe (dB)
25.17 23.27 27.59
51.45 47.95 56.56
SNRe 5 elastographic signal-to-noise ratio; CNRe 5 elastographic contrast-to-noise ratio.
To assess the effect of compounding on strain image quality under different amounts of compression, multiple consecutive image frames were acquired while compression was being applied mechanically. Every pair of two successive image frames was made to be different in the
a
29
28
SNRe [dB]
27
26
25
24
23 0.2
0.4
0.6
0.8
1
1.2
1.4
1
1.2
1.4
strain [% ]
b
58
56
CNRe [dB]
54
52
50
48
46 0.2
0.4
0.6
0.8 strain [% ]
Fig. 8. Comparison of the (a) elastographic signal-to-noise ratio (SNRe) and (b) elastographic contrast-to-noise ratio (CNRe) values as the amount of strain applied is increased. The solid and dotted lines represent the cases with and without compounding, respectively. The error bars denote one standard deviation.
Volume 36, Number 4, 2010
amount of compression by 0.42%. Images that are two frames apart were taken to make the amount of compression equal to 0.84% and their strain was estimated to construct strain image. Also, images that are three frames apart were taken to make the amount of compression equal to 1.26% and their strain is estimated to construct strain image. Figure 8 shows the effect of compounding as the applied strain is variably set to 0.42%, 0.84%, and 1.26%. The dotted lines represent the case of using either of the two array transducers while the solid line represents the case of compounding both array transducers. Figures 8a and b plot the SNRe and CNRe values, respectively, as the applied strain is increased. The vertical error bars denote one standard deviation from the mean. The SNRe value of the compounded strain image has increased by more than 2.5 dB compared with the uncompounded strain images. It is known that in general, increasing the amount of applied strain to within some range tends to improve the SNRe value. It is found, however, that compounding of strain images is more effective in reducing noise (Lubinski et al. 1999). The strain image quality can also be improved by averaging over consecutive image frames based on persistence. Persistence is a signal processing method widely used to improve ultrasound image quality by averaging consecutive moving image frames and thereby reducing noise (Wang and Liu 2006). In the method, the noise can be suppressed considerably but the spatial resolution is degraded because the constituent strain image frames do not exactly match geometrically. Those image frames need to be geometrically matched using an affine transform. In general, in a manual palpation method, the amount of strain applied cannot be exactly estimated so the geometrical distortion cannot be compensated for exactly. On the contrary, in our proposed method, the amount of strain applied is exactly known and, thus, the distortion can be compensated for. Figure 9 plots the strain profiles along the scan lines that pass through the center of the cylinder in frames 1, 6, and 11 of strain images which are obtained successively, as a constant strain of 0.42% is applied from one frame to the next in all steps of the data acquisition process. The strain between frames 1 and 2 is 0.42%, that between frames 1 and 3 is 0.84%, that between frames 1 and 4 is 1.26%, and so on. The strain increases with increasing frame number. The strain values inside the cylinder are shown to be small. However, neither the falling nor the rising edges shown in Figure 9a coincide exactly. Figure 9b shows the result of performing a global stretching operation in Figure 9a using the applied strain where both left and right edges are seen to be aligned. Averaging these aligned strain images results in a decrease of noise without degrading the image resolution.
Multimode ultrasound breast imaging d M.-K. JEONG and S.-J. KWON
645
1.8 1.6
relative magnitude
1.4 1.2 1 0.8 0.6 0.4 0.2 0
0
50
100
150
200
250
pixel
Fig. 10. Strain image after global stretching is applied to RF data. The cylindrical inclusion is of diameter 10 mm.
1.8 1.6
relative magnitude
1.4 1.2 1 0.8 0.6 0.4 0.2 0
0
50
100
150
200
250
pixel
Fig. 9. Strain profiles along the scan lines that pass through the center of the cylinder in frames 1, 6 and 11 of consecutively obtained strain images by increasing the applied strain in steps of 0.42% with increasing frame number without (top) and with (bottom) global stretching.
suitable for imaging protruding objects such as a human breast placed between two opposing linear array transducers. In vitro images of compounding and SOS as well as strain are reconstructed. Although there is some distortion in the SOS images due to limited view angle, those three types of images can be used complementarily to aid in better diagnosis. Unlike computed or diffraction tomography that uses water to surround a breast, our imaging method is advantageous in that there is no need to correct for refraction at the surface of it since the transducers are in direct contact with it. The surface refraction is one of the major sources of error in SOS estimation. The data required for the proposed method can be acquired very fast if a dedicated hardware system that operates in real time is built. Therefore, the diagnosis can be made within a small amount of time without causing much discomfort to the patient. Acknowledgements—This work was supported by Medison.
Figure 10 shows the result of applying global stretching to the RF data where the applied strain is 0.42%. The diameter of the cylindrical inclusion is 10 mm. The soft region is stretched to equal its original length, but the hard region is expanded because the latter has not been compressed much. Thus, the black and white levels are reversed so that the hard region appears to be brighter. This characteristic helps us better identify the hard region. Our method has the advantage that the amount of the total compression is available. CONCLUSIONS We have investigated a new linear array transducer configuration for compound, SOS, and strain imaging. Its feasibility was confirmed by experiments with two types of in-house phantoms. The proposed method is
REFERENCES Bilgen M, Insana MF. Predicting target detectability in acoustic elastography. Proc IEEE Ultrason Symp 1997;1427–1430. Carson PL, Meyer CR, Scherzinger AL, Oughton TV. Breast imaging in coronal planes with simultaneous pulse echo and transmission ultrasound. Science 1981;214:1141–1143. Cespedes I, Ophir J. Reduction of image noise in elastography. Ultrason Imag 1993;15:89–102. Duric N, Li C, Glide-Hurst C, Littrup P, Huang L, Lupinacci J, Schmidt S, Rama O, Bey-Knight L, Xu Y. Breast imaging with ultrasound tomography: Clinical results at the Karmanos Cancer Institute. 2008 Int Conf Biomed Eng Informatics 2008;713–717. Erkamp RQ, Wiggins P, Skovoroda AR, Emelianov SY, O’Donnell M. Measuring the elastic modulus of small tissue samples. Ultrason Imag 1998;20:17–28. Glover GH, Sharp JC. Reconstruction of ultrasound propagation speed distributions in soft tissue: Time-of-flight tomography. IEEE Trans Sonic Ultrason 1977;24:229–234. Goss SA, Johnston RL, Dunn F. Comprehensive compilation of empirical ultrasonic properties of mammalian tissues. J Acoust Soc Amer 1978;64:423–457.
646
Ultrasound in Medicine and Biology
Greenleaf JF, Johnson SA, Samoyoa WF, Duck FA. Algebraic reconstruction of spatial distributions of acoustic velocities in tissue from their time-of-flight profiles. Acoust Holography 1975;6:71–90. Greenleaf JF, Bahn RC. Clinical imaging with transmissive ultrasonic computerized tomography. IEEE Trans Biomed Eng 1981;28:177–185. Hall TJ. The physics of elasticity imaging: A new option on the latest US systems. Proc RSNA 2004;233–246. Hansen C, Schasse A, Hu¨tterbra¨uker N, Ashfaq M, Wilkening W, Ermert H. Reconstruction of speed of sound for a correction of transit time in full angle spatial compounding. Proc IEEE Ultrason Symp 2007;785–788. Huang S-W, Li P-C. Computed tomography sound velocity reconstruction using incomplete data. IEEE Trans Ultrason Ferroelectr Freq Control 2004;51. 329–243. Jeong M-K, Kwon S-J, Cho S-M, Bae M-H, Kim Y-G. Ultrasound breast imaging technique using two opposing transducer arrays. Proc IEEE Ultrason Symp 2008;1274–1277. Konofagou EE, Ophir J, Kaller F, Varghese T. Elastographic dynamic range expansion using variable applied strains. Ultrason Imag 1997;19:145–166. Krueger M, Pesavento A, Ermert H. A modified time-of-flight tomography concept for ultrasonic breast imaging. Proc IEEE Ultrason Symp 1996;1381–1385. Kru¨cker JF, Fowlkes JB, Carson PL. Sound speed estimation using automatic ultrasound image registration. IEEE Trans Ultrason Ferroelectr Freq Control 2004;51:1095–1106. Li C, Duric N, Huang L. Breast imaging using transmission ultrasound: Reconstructing tissue parameters of sound speed and attenuation. 2008 Int Conf Biomed Eng Informatics 2008;708–712. Li C, Duric N, Littrup P, Huang L. In vivo breast sound-speed imaging with ultrasound tomography. Ultrasound Med Biol 2009;35: 1615–1628. Lopata RGP, Nillesen MM, Hendrik HGH, Gerrits IH, Thijssen JM, de Korte CL. Performance evaluation of methods for two-dimensional
Volume 36, Number 4, 2010 displacement and strain estimation using ultrasound radio frequency data. Ultrasound Med Biol 2009;35:796–812. Lubinski MA, Emelianov SY, O’Donnell M. Adaptive strain estimation using retrospective processing. IEEE Trans Ultrason Ferroelectr Freq Control 1999;46:97–107. Mizutani K, Nishizaki K, Nagai K, Harakawa K. Measurement of temperature distribution in space using ultrasound computerized tomography. Jpn J Appl Phys 1997;36:3176–3177. Ophir J, Ce´spedes I, Ponnekanti H, Yazdi Y, Li X. Elastography: A quantitative method for imaging the elasticity of biological tissues. Ultrason Imag 1991;13:111–134. Robinson DE. Digital reconstruction and display of compound scan ultrasound images. IEEE Trans Sonics Ultrason 1984;31: 396–406. Robinson DE, Ophir J, Wilson LS, Chen CF. Pulse-echo ultrasound speed measurements: Progress and prospects. Ultrasound Med Biol 1991;17:633–646. Sarvazyan AP, Rudenko OV, Swanson SD, Fowlkes JB, Emelianov SY. Shear wave elasticity imaging: A new ultrasonic technology of medical diagnostics. Ultrasound Med Biol 1998;24:1419–1435. Techavipoo U, Varghese T. Improvements in elastographic contrast-tonoise ratio using spatial-angular compounding. Ultrasound Med Biol 2005;31:529–536. Varghese T, Ophir J. Enhancement of echo-signal correlation in elastography using temporal stretching. IEEE Trans Ultrason Ferroelectr Freq Control 1997;44:173–180. Wang G, Liu DC. Adaptive persistence utilizing motion compensation for ultrasound images. Proc 18th Int Conf Pattern Recognition 2006;3:726–729. Wilhjelm JE, Jensen MS, Brandt T, Sahl B, Martinsen K, Jespersen SK, Falk E. Some imaging strategies in multi-angle spatial compounding. Proc IEEE Ultrason Symp 2000;1615–1618. Yoon R-Y, Kwon S-J, Bae M-H, Jeong M-K. Improved ultrasonic elasticity imaging with center frequency estimation and global shift compensation. Proc IEEE Ultrason Symp 2006;1278–1281.