Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Sommer, G.
Right arrow Articles by Tai, L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Sommer, G.
Right arrow Articles by Tai, L.
(Radiology. 1999;211:629-636.)
© RSNA, 1999

Ultrasonography

Liver Tumors: Utility of Characterization at Dual-Frequency US1

Graham Sommer, MD, Eric W. Olcott, MD and Lisa Tai, MD

1 From the Departments of Radiology (G.S.) and Surgical Pathology (L.T.), Stanford University Medical Center, 300 Pasteur Dr, Rm H-1307, Stanford, CA 94305-5621, and the Department of Radiology, Veterans Affairs Palo Alto Care System, Calif (E.W.O.). Received May 22, 1998; revision requested July 14; revision received October 14; accepted December 16. Address reprint requests to G.S.


   Abstract
TOP
 Abstract
Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
PURPOSE: To evaluate the potential utility of a technique for analysis of ultrasonographic (US) images obtained at two center frequencies in a phantom model and in a clinical study of liver tumors.

MATERIALS AND METHODS: A US phantom was built that contained target regions with polystyrene scatterers both smaller and larger than scatterers in the background material. High- and low-frequency US images of target regions were obtained, and changes in the contrast-to-noise ratio (CNR) of the targets relative to the background were determined. In a clinical evaluation, 12 hemangiomas, 13 liver metastases, one adenoma, and two hepatomas were evaluated in a similar manner, and the results were correlated with the histologic findings in resected lesions.

RESULTS: For targets containing scatterers smaller than those in the background, there was an increase in CNR of higher frequency images. The converse was true for target regions having scatterers larger than those in the background. Liver metastases generally showed an increase in CNR of higher frequency images, whereas hemangiomas generally showed somewhat decreased CNR of higher frequency images, differing significantly from metastases (P < .01–.001).

CONCLUSION: Changes in CNRs of relatively high- and low-frequency US images may give useful information on the frequency dependence of backscattering, which is descriptive of the histologic findings in lesions and which is not evident with conventional US images.

Index terms: Angioma, gastrointestinal tract, 761.12989, 761.3194 • Liver neoplasms, US, 761.12989, 761.3192, 761.3194, 761.33 • Ultrasound (US), experimental studies, 761.12989 • Ultrasound (US), tissue characterization, 761.12989


   Introduction
TOP
 Abstract
 Introduction
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 
The intensity of ultrasound scattered in tissue is highly frequency dependent, and it is well known that this frequency dependence of ultrasound scattering is dependent on the shape, elastic properties, and size of the scatterers present in the tissue (1). The ratio of backscattered to incident ultrasound may be expressed as a "backscattering cross section" (2) and Sb = Io x {sigma}b(f), where Sb and I0 are the backscattered and incident ultrasound intensities, respectively, and {sigma}b(f) is the backscattering cross section, which is a function of frequency, f.

In present-day ultrasonographic (US) imaging, images are created at a single frequency, and information on scattering frequency dependence, which might provide important indications as to the histologic features of tissues, is not obtained. In prior work, investigators have been successful in characterizing ocular tumors (3), scattering structures within the kidney (4,5), and diffuse liver disease (6) by creating US images that encode regional estimates of the size of tissue scatterers. These techniques were based on the characteristics of the frequency spectra of digitized backscattered ultrasound waveforms. The purpose of the present study was to evaluate a different approach to obtaining information on scatterer sizes in tissues: US images were created at two center frequencies to estimate the frequency dependence of backscattering in focal lesions compared with that in the background tissues in a phantom model and in a clinical study of liver tumors.


   MATERIALS AND METHODS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 References
 
A US phantom was constructed with overall dimensions 6 x 11 x 18 cm. The general design was of a background of agar material (Sigma Chemical, St Louis, Mo) and three target regions with different size ranges of polystyrene divinylbenzene microspheres (Duke Scientific, Palo Alto, Calif) suspended in the agar gel. Both the background gel and the target regions were created in the proportions of 2 g of polystyrene microspheres and 15 g of dry agar to 1 L of boiling water. All mixtures were boiled for 5 minutes and poured into the appropriate target molds or in the larger container enclosing the overall phantom.

Three target regions were created by using polystyrene divinylbenzene microspheres, each having a different microsphere size range: 1–85 µm, 50–130 µm, and 250–750 µm. The background gel was created by using 100–500-µm microspheres. The three target regions were created by pouring the three gel mixtures into latex condoms to form cylindric target regions 3 cm in diameter and 6 cm long. These were then suspended in a plastic container that had the same dimensions as the overall phantom, at a depth of 8 cm from the phantom surface, and the background material was poured. Both the overall phantom and the individual target regions were rotated continually in an ice bath during hardening to prevent settling of the microspheres. The phantom was then removed from the box for use in the following studies.

US images of the three target regions within the phantom were obtained by using a US machine (Sequoia; Acuson, Mountain View, Calif) and three different US probes capable of creating images with the indicated minimum and maximum center frequencies: sector probe, 2.5 and 4.0 MHz; convex probe, 2.5 and 5.0 MHz; and convex probe, 5.0 and 8.0 MHz. With the individual probes held in a stand, images of all target regions were obtained at both the maximum and minimum center frequencies available, and overall and time-dependent gain were carefully adjusted to maintain a homogeneous overall US image in each case. Image files in PICT format were created by using an Aegis Mini-PACS System (Acuson), and all subsequent image analysis and processing was performed after importation into the Photoshop 4.0 (Adobe Systems, Mountain View, Calif) program.

Contrast-to-noise ratios (CNRs) for each of the targets within the background were created by using regions of interest within the target regions themselves and within similar regions in the background material at exactly the same depth ranges as in the phantom, as illustrated in Figure 1 part A. CNRs were computed by using the formula CNR = (AL - AB)/({sigma}L2 + {sigma}B2)1/2, where AL and AB are the mean amplitudes of the target lesion and background, respectively, and {sigma}L and {sigma}B are the SDs of the amplitudes of the target lesion and background, respectively. The characteristics of the transducer are proprietary, and the influence of these characteristics on the results was not evaluated.



View larger version (60K):
[in this window]
[in a new window]
  		Figure 1. US images of the small-scatterer (1–85-µm microspheres) phantom target obtained at 5.0- and 8.0-MHz center frequencies. A, Image obtained at 5.0 MHz shows the regions of interest (rectangles) in the target and in the adjacent phantom background (containing 100–500-µm microspheres) used to calculate CNR. B, US image obtained at 5.0 MHz and C, US image obtained at 8.0 MHz show a substantial increase in echogenicity of the target (arrow in B) relative to the background on the 8.0-MHz image. D, Image obtained by direct subtraction of B from C. E, Image obtained by subtraction of B from C after low-pass filtration of the two images. F, Subtracted image of E used as a color overlay on C.

 
The CNR changes between the high- and low-frequency images for target regions relative to the background were then computed for each of the individual targets by subtracting the CNRs of the targets at lower frequencies from the CNRs at higher frequencies for each of the three probes. To create single images encoding the change in CNR, direct subtraction of the images obtained at high and low frequencies was performed, before and after low-pass filtering to reduce ultrasound speckle noise, to provide a pictorial representation of regions of substantial change in CNR.

After approval by the institutional review board of our institution (Stanford University School of Medicine), written informed consent was obtained prior to obtaining US scans of patients with known liver lesions in a clinical evaluation of the dual-frequency technique. Dual-frequency images were obtained for 12 hemangiomas in 10 patients (seven women, three men; age range, 28–61 years); the hemangiomas were proved by the typical appearance of the lesions on computed tomographic (CT) or magnetic resonance (MR) images (7–9) in all patients, except for one 28-year-old woman for whom stable size and appearance of the hemangiomas were documented from US scans obtained 4 months earlier.

Thirteen metastases in nine patients and one adenoma and two hepatomas in individual patients (10 men, two women; age range, 37–81 years) were also studied. The metastases were confirmed by the typical appearance on CT scans in patients with known primary tumors—colon tumor, three; non-small cell lung carcinoma, one; carcinoid, one; gastrointestinal stromal tumor, one; pancreatic adenocarcinoma, one; and renal cell carcinoma, one—and one patient with biopsy-proved adenocarcinoma for which the primary tumor was unknown. The adenoma and hepatomas were proved histologically following resection. US images were obtained by using the same US machine and probes as described for the phantom study.

The change in CNR values of the individual liver lesions was computed in exactly the same manner as that described for the phantom; CNR computations for two to five image pairs were generally averaged to minimize spatial sampling errors. Care was taken to use regions of interest that avoided regions of acoustic shadowing and to sample the most representative regions of lesions that appeared heterogeneous. Prior to quantitative analysis of the image pairs, images that were deemed technically suboptimal or image pairs that appeared somewhat spatially misregistered were excluded from our computations. In four patients with liver metastases, the change in CNR for two discrete lesions was averaged. The change in CNR values for the individual probes was then tabulated according to lesion type, and the significance of the change in CNR for the differentiation of liver metastases from hemangiomas was determined by using a two-tailed Student t test for each of the three probes. A P value less than .01 was considered to indicate a statistically significant difference.

Three patients who underwent scanning subsequently underwent partial hepatectomy for the removal of lesions (one metastasis from adenocarcinoma of the colon, one adenoma, and one hepatoma) imaged by using the dual-frequency technique. Representative portions of these tumors, as well as normal liver and a typical hemangioma, were stained with Masson trichrome stain and imaged with low-power microscopy. The purpose of this portion of the study was to determine whether the scattering characteristics of the lesions might explain the quantitative results of the dual-frequency analysis.


   RESULTS
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 References
 
Phantom Study
Representative images of the US phantom that show a small-scatterer target (scatterer size range, 1–85 µm) obtained with the convex 5.0–8.0-MHz probe are shown in Figure 1, and images of the larger scatterer target (scatterer size range, 250–750 µm) obtained with the sector 2.5–4.0-MHz probe are shown in Figure 2. There was a substantial change in the CNRs of the two targets relative to the background at different frequencies, and the small-scatterer target became brighter at a higher frequency (the CNR was -1.34 at 5.0 MHz and increased to -0.26 at 8.0 MHz), with a positive change in CNR of 1.08, whereas the larger scatterer target was brighter at a lower frequency (the CNR was 0.40 at 2.5 MHz and decreased to -0.32 at 4.0 MHz), with a negative change in CNR of -0.72.



View larger version (45K):
[in this window]
[in a new window]
  		Figure 2. US images of the larger scatterer (250–750-µm microspheres) phantom target obtained at 2.5 and 4.0 MHz. A, Image obtained at 2.5 MHz shows that the target region is difficult to detect but is slightly hyperechoic relative to the background (containing 100–500-µm microspheres). B, Image obtained at 4.0 MHz shows that the target (arrow) becomes slightly hypoechoic. C, Image created with a similar processing technique to that used in Figure 1 part F, except that the higher frequency image was subtracted from the lower frequency image after low-pass filtration of the two images. The red overlay shows the spatial distribution of scatterers of relatively low frequency dependence of scattering compared with that of the background.

 
The changes in CNR values for the three targets are summarized in Figure 3. For the two targets with scatterers smaller than those in the background, the change in CNR was positive, and the change in CNR increased with higher probe center frequencies. Conversely, for the large-scatterer target, the change in CNR was negative for all probes, except for the convex 5.0–8.0-MHz probe.



View larger version (13K):
[in this window]
[in a new window]
  		Figure 3. Bar graph summarizes the calculated change in CNR for the three phantom targets, two of which contained scatterers smaller than those in the background, which contained 100–500-µm microspheres, and one of which contained scatterers larger than those in the background, for the three US probes.

 
Figure 1 parts D and E show that it was possible to represent spatially the regions that changed substantially in CNR between the high- and low-frequency images by creating a simple subtracted image of the high-frequency image minus the low-frequency image following standardization of the mean and SD of the background gray levels in the two images. Figure 1 part E, obtained after low-pass filtration of the two source images, is much less noisy, however, because the subtraction of the uncorrelated speckle noise in the two original source images produced an undesirably noisy resultant image (Fig 1 part D). Figure 1 part F shows that it was possible to use the results of subtraction after low-pass filtration to create a color overlay that provided spatial depiction of regions of positive change in CNR.

Conversely, Figure 2 shows images of the phantom target with relatively large scatterers (250–750 µm) compared with those in the background; here, the target appears brighter on the lower frequency 2.5-MHz image (Fig 2 part A) than on the higher frequency 4.0-MHz image (Fig 2 part B), with a resultant decrease in CNR. The subtraction of the high- from the low-frequency image after low-pass filtration allowed the creation of a red overlay that indicated the spatial distribution of regions of negative change in CNR (Fig 2 part C).

Clinical Study of Liver Tumors
Most hemangiomas were somewhat brighter on the lower frequency images than on the higher frequency images, as shown in Figure 4. This resulted in negative calculated values of change in CNR in most cases, which was similar to findings with the large-scatterer phantom target. Figure 5 shows representative lower and higher frequency images for three liver metastases and a hepatoma. For most of the liver metastases, as well as the hepatomas and adenoma, the lesions appeared somewhat brighter at higher frequencies, with resultant positive calculated changes in CNR values, which was analogous to the small-scatterer phantom targets.



View larger version (78K):
[in this window]
[in a new window]
  		Figure 4a. Representative clinical US images of liver hemangiomas obtained at lower and higher frequencies. The hemangiomas are indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a hemangioma in the left lobe of the liver in a 60-year-old woman; the measured change in CNR is -0.30. (b) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of a smaller hemangioma in the right lobe of the liver in a 30-year-old woman; the measured change in CNR is -0.14. The negative change in CNR values implies that the scatterer size is larger in the hemangiomas than in the adjacent liver parenchyma.

 


View larger version (69K):
[in this window]
[in a new window]
  		Figure 4b. Representative clinical US images of liver hemangiomas obtained at lower and higher frequencies. The hemangiomas are indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a hemangioma in the left lobe of the liver in a 60-year-old woman; the measured change in CNR is -0.30. (b) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of a smaller hemangioma in the right lobe of the liver in a 30-year-old woman; the measured change in CNR is -0.14. The negative change in CNR values implies that the scatterer size is larger in the hemangiomas than in the adjacent liver parenchyma.

 


View larger version (78K):
[in this window]
[in a new window]
  		Figure 5a. Lower and higher frequency US images of four malignant liver lesions, indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a liver metastasis from adenocarcinoma of the colon in a 71-year-old man; the measured change in CNR is +0.27. (b) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from renal cell carcinoma in a 58-year-old man; the measured change in CNR is +0.43. (c) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from colonic adenocarcinoma in a 37-year-old man; the measured change in CNR is -0.06. A histologic image of this lesion is shown in Figure 7 part D. (d) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of hepatoma in a 42-year-old man; the measured change in CNR is +0.85. A histologic image of this lesion is shown in Figure 7 part B. The positive change in CNR values for these lesions, with the exception of that in c, implies that the lesion scatterer sizes are smaller than those in the adjacent hepatic parenchyma.

 


View larger version (69K):
[in this window]
[in a new window]
  		Figure 5b. Lower and higher frequency US images of four malignant liver lesions, indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a liver metastasis from adenocarcinoma of the colon in a 71-year-old man; the measured change in CNR is +0.27. (b) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from renal cell carcinoma in a 58-year-old man; the measured change in CNR is +0.43. (c) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from colonic adenocarcinoma in a 37-year-old man; the measured change in CNR is -0.06. A histologic image of this lesion is shown in Figure 7 part D. (d) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of hepatoma in a 42-year-old man; the measured change in CNR is +0.85. A histologic image of this lesion is shown in Figure 7 part B. The positive change in CNR values for these lesions, with the exception of that in c, implies that the lesion scatterer sizes are smaller than those in the adjacent hepatic parenchyma.

 


View larger version (93K):
[in this window]
[in a new window]
  		Figure 5c. Lower and higher frequency US images of four malignant liver lesions, indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a liver metastasis from adenocarcinoma of the colon in a 71-year-old man; the measured change in CNR is +0.27. (b) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from renal cell carcinoma in a 58-year-old man; the measured change in CNR is +0.43. (c) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from colonic adenocarcinoma in a 37-year-old man; the measured change in CNR is -0.06. A histologic image of this lesion is shown in Figure 7 part D. (d) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of hepatoma in a 42-year-old man; the measured change in CNR is +0.85. A histologic image of this lesion is shown in Figure 7 part B. The positive change in CNR values for these lesions, with the exception of that in c, implies that the lesion scatterer sizes are smaller than those in the adjacent hepatic parenchyma.

 


View larger version (99K):
[in this window]
[in a new window]
  		Figure 5d. Lower and higher frequency US images of four malignant liver lesions, indicated with arrows on the 2.5-MHz images. (a) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of a liver metastasis from adenocarcinoma of the colon in a 71-year-old man; the measured change in CNR is +0.27. (b) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from renal cell carcinoma in a 58-year-old man; the measured change in CNR is +0.43. (c) Transverse images obtained at 2.5 MHz (left) and 5.0 MHz (right) of metastasis from colonic adenocarcinoma in a 37-year-old man; the measured change in CNR is -0.06. A histologic image of this lesion is shown in Figure 7 part D. (d) Sagittal images obtained at 2.5 MHz (left) and 4.0 MHz (right) of hepatoma in a 42-year-old man; the measured change in CNR is +0.85. A histologic image of this lesion is shown in Figure 7 part B. The positive change in CNR values for these lesions, with the exception of that in c, implies that the lesion scatterer sizes are smaller than those in the adjacent hepatic parenchyma.

 
Using a two-tailed t test, we found that the trend for a lower change in CNR in hemangiomas compared with the change in CNR in metastases was statistically significant (P < .01) for the convex 5.0–8.0-MHz probe, for which a lack of adequate penetration prevented evaluation of many lesions (Fig 6). The differentiation of hemangiomas from metastases was statistically significant for the sector 2.5–4.0-MHz probe and the convex 2.5–5.0-MHz probe (P < .001 for both). There was some degree of overlap, however, with two hemangiomas that showed a positive change in CNR with each of the two latter probes and one liver metastasis that showed a negative change in CNR with the convex 2.5–5.0-MHz probe.



View larger version (8K):
[in this window]
[in a new window]
  		Figure 6. Plot of CNR changes measured with the three probes used in this study for the liver tumors evaluated. Most of the hemangiomas have a negative change in CNR values, and most hepatomas, metastases, and adenomas have a positive change in CNR values. {circ} = hemangiomas, • = metastases, {triangleup} = adenomas, {blacktriangleup} = hepatomas.

 
Histologic images from the three patients who underwent partial hepatectomy after dual-frequency US scanning are shown in Figure 7, along with a histologic image of a typical liver hemangioma. All photomicrographs were obtained at the same magnification, following trichrome staining, which stains collagen an intense blue color. In the normal liver, collagen is seen primarily in the normal portal triad regions. The size of the portal tracts and the luminal structures contained within them vary within the liver according to their proximity to the hepatic hilum (10); in the photomicrograph of normal liver shown in Figure 7 part A, the average width of the portal tracts was 250 µm.



View larger version (143K):
[in this window]
[in a new window]
  		Figure 7. Histologic images of normal liver and liver tumors stained with trichrome stain and imaged at exactly the same low-power microscopic settings. (Masson trichrome stain; original magnification, x30.) The length of the field is 3.25 mm; the length of 1.0 mm is indicated in A. A, Typical appearance of normal liver resected from a 37-year-old man in this study. The trichrome stain gives collagen an intense blue appearance, as seen in the approximately 250-µm-diameter portal triads (arrows). B, Central portion of a hepatoma (Fig 5d) imaged in this study in a 42-year-old man. The tumor is highly cellular with only thin strands of collagen throughout. C, Central portion of an adenoma evaluated in this study in a 38-year-old woman. No collagen is visible, and the tumor is highly cellular, with prominent glycogen deposition in the hepatocytes. D, Metastasis from colonic adenocarcinoma imaged in this study in a 37-year-old man (Fig 5c). There is a marked collagenous response with broad bands of blue-staining collagen throughout the central portions of this tumor. E, Typical appearance of a hemangioma obtained in a 54-year-old woman. A great deal of collagen is present in the supporting structure of the hemangioma. The vascular channels are seen as clear areas that contain some residual red cells.

 
There were substantial differences in the amount and distribution of collagen in the four liver tumors represented: The nonfibrolamellar hepatoma (Fig 7 part B) was predominantly cellular, with very thin strands of collagen throughout that represent the network of sinusoidal vessels surrounding the tumor cells. The adenoma (Fig 7 part C) contained no discernible collagen but had prominent glycogen deposition within the hepatocytes. The colonic adenocarcinoma metastasis represented in Figure 7 part D showed abundant large regions of collagen deposition incited by the metastatic tumor; the hemangioma (Fig 7 part E) also had abundant collagen staining but in a different distribution, with abundant collagen surrounding endothelium-lined, nonanastomotic vascular channels.


   DISCUSSION
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
References
 
Physical Principles of Dual-Frequency US
As noted in the Introduction, the dual-frequency technique of lesion characterization used in this study is based on differences in backscatter frequency dependence in tissues, which is in turn largely dependent on the relationship between the insonating wavelengths and the scatterer sizes. It is the size of the insonating ultrasound wavelength relative to the size of the scatterers that is the critical factor in determining the intensity of backscattered ultrasound. If the wavelength is small compared with the dimensions of the scatterer, the reflection will be specular and approximately frequency independent (f0); if the wavelength is much larger than that of the scattering center, the intensity of backscattered ultrasound will be a strong function of frequency (f4 dependency, Rayleigh scattering). For wavelengths intermediate between these two, there will be a progressive increase in backscatter frequency dependence as the wavelength becomes greater relative to the scatterer size.

The general principle behind the dual-frequency US technique is illustrated diagrammatically in Figure 8, in which insonation at a higher and a lower frequency of a tissue containing lesions with scatterers both larger and smaller than the host tissue is shown. There is more high-frequency ultrasound scattered from the lesion with scatterers smaller than those in the host tissue owing to their higher scattering frequency dependence, and for the opposite reason, more low-frequency ultrasound is scattered from the lesion with larger scatterers.



View larger version (54K):
[in this window]
[in a new window]
  		Figure 8. Diagrammatic representation of the general principle that allows the differentiation of lesions that contain either larger or smaller scatterers than those in the background by means of the dual-frequency technique. The thicker arrows represent backscattered ultrasound of relatively high intensity. Insonation of tissues is performed at two frequencies; normalization of the background intensity ensures that the background will have similar amplitude on low- and high-frequency images. The target that contains scatterers larger than those in the background will preferentially backscatter a higher proportion of low-frequency ultrasound owing to a lower frequency dependence of backscatter than that of the background tissue, and the target region with smaller scatterers will preferentially scatter at higher frequencies owing to a higher frequency dependence of backscatter. The result is that the large-scatterer target will appear brighter than the background at lower frequencies, and the small-scatterer target will appear brighter than the background at higher frequencies. This phenomenon explains the change in CNR values obtained for the phantom and clinical studies performed.

 
Phantom Studies
The general principles of dual-frequency US outlined earlier can be illustrated with the results of the phantom study. An example of imaging a region with scatterers smaller than those in the background is illustrated in Figure 1, which depicts the target region containing 1–85-µm polystyrene microspheres in the background of 100–500-µm microspheres. On the 5.0-MHz center frequency image (Fig 1 part B), the lesion appeared hypoechoic, but the relative amplitude of echoes in the target relative to the background became much greater at 8.0 MHz (Fig 1 part C) and the target region appeared almost isoechoic.

The mechanism for this observation is related to the differing ultrasound wavelengths at 5.0 MHz (300 µm) and 8.0 MHz (186 µm) and the scatterer sizes. Because the background scatterers are close in size to these two wavelengths, there will be little increase in their backscatter at the higher 8.0-MHz frequency; but there will be much more backscatter from the target at the higher frequency, because it contains scatterers smaller than either wavelength and will exhibit high scattering frequency dependence. This will result in the target appearing substantially brighter relative to the background at the higher frequency.

An example of the opposite situation, imaging a target region with larger scatterers than those in the background, is shown in Figure 2. In this case, the target region appeared hyperechoic at the lower frequency and hypoechoic at the higher frequency, owing to the same general physical mechanisms described for Figure 1. It is evident from Figures 1 and 2 that the terms "hyperechoic" and "hypoechoic" are dependent on the center frequency of insonation. The color image overlays were created by means of image subtraction to investigate the possibility that such image processing could indicate the frequency dependence of a lesion relative to the background in a single image (Figs 1 part F, 2 part C).

Application to Characterization of Liver Tumors
The clinical evaluation of liver tumors indicated that the technique of dual-frequency US may be useful in the characterization of liver tumors, if one bears in mind the limitation that only imaging and clinical features were available for the verification of lesion types. Hemangiomas evaluated by means of this technique showed a change in CNR that was generally negative. This general appearance is noted in Figure 4, in which two representative hemangiomas both appeared somewhat more echogenic on the lower frequency image than on the higher frequency image. Conversely, most liver metastases appeared more echogenic than the background on the higher frequency image, and thus had positive measured changes in CNR. In Figure 5a and 5b, a hypoechoic metastasis and a hyperechoic metastasis demonstrate this general effect. An incremental ability over conventional US to differentiate hemangiomas from liver metastases could be beneficial, because such differentiation is a common clinical problem.

Hemangiomas are very common liver tumors that occur in about 7% of the general population (11). Although they often look like well-defined hyperechoic lesions on US images, the differentiation from metastases, particularly hyperechoic metastases, is often difficult and necessitates the performance of contrast material–enhanced CT scanning or MR imaging (7–9). Insight into the physical reasons for the findings of the study of dual-frequency US in liver tumors may be evident from the evaluation of the histologic findings in the normal liver and in liver tumors.

Figure 7 shows representative central portions of three resected tumors (a hepatoma, a hepatic adenoma, and a colonic metastasis) and normal liver and a typical hemangioma. In the specimen of normal liver, there were a number of blue-staining portal triads, averaging about 250 µm each in diameter (Fig 7 part A), and these can be expected to be the major scatterers in normal liver. In Figure 7 part B, a representative section of the central portion of a hepatoma, there are only thin bands of collagen in this highly cellular tumor. This is the tumor that is imaged in the dual-frequency US scans shown in Figure 5d, in which central portions of the tumor were substantially brighter at the higher frequency and showed a positive change in CNR. The tumor was thus behaving like the phantom target with smaller scatterers than those in the background; similarly, the hepatoma contained collagen organized on a smaller scale than in the normal liver, and the findings can be explained on the basis of differences in scattering frequency dependence.

While many metastases can be expected to be highly cellular, similar to the adenoma in Figure 7 part C, the metastasis shown in Figure 7 part D was characterized by abundant collagen deposition in broad bands. This is a colonic metastasis that incited an intense collagenous response, which led to the appearance shown. In the dual-frequency study of liver tumors, this metastasis behaved somewhat differently than the other metastases, which generally appeared more echogenic at higher frequencies; as shown in Figure 5c, the lesion was virtually isoechoic at both lower and higher frequencies, which is presumably related to the large-scale collagen deposition (12).

Figure 7 part E depicts a typical hemangioma of the liver, which shows dense blue-staining collagenous tissue surrounding the walls of the vascular channels in the tumor. The hemangioma thus had a distinctly different appearance in terms of amount and distribution of collagen than the highly cellular hepatoma and adenoma in Figure 7 parts B and C, respectively. The larger amount of collagen, compared with that found in normal liver, most likely leads to a lower frequency dependence of backscattering and explains the tendency for a negative change in CNR measurements with dual-frequency US.

Prior Related Work
In prior studies (13,14) in which narrow-band frequency filtration of ultrasound was used, it was possible to alter the contrast of lesions in their surroundings, owing to the same physical principles discussed earlier for the dual-frequency technique. There have also been a number of prior attempts (1,3–6,15) to exploit frequency-dependent backscattering effects to perform regional estimates of the sizes of scatterers in tissue and thus obtain information incremental to that of conventional US imaging. These techniques have generally involved digitization of ultrasound waveforms backscattered from tissues of interest prior to computation of local frequency spectra. Features of these spectra were then used to perform local estimates of scatterer size and often other features of the tissues insonated.

Some recent study results are supportive of the general observations made in this study. In a study of liver metastases and hemangiomas (16) in which radio-frequency digitization of backscattered ultrasound was used, it was found that the mean frequency of ultrasound backscattered from metastases is somewhat higher than that from hemangiomas, compared with that from adjacent normal liver, a result in accord with the findings of the present study. In a recent comprehensive quantitative US analysis of liver metastases, Huisman et al (17) digitized ultrasound waveforms from a large number of liver metastases and adjacent normal liver parenchyma. Using selective frequency filtration, they obtained US images that corresponded to a range of center frequencies, which indicated that the backscattering frequency dependence in hypoechoic metastases is greater than that of adjacent normal liver tissue, a finding also in accord with the findings of the present study.

Limitations and Preferred Embodiment of the Technique
Although the technique of dual-frequency US used in this study shows promise for the characterization of liver tumors, and in particular the differentiation of liver metastases from hemangiomas, there are a number of difficulties with its application. Because the technique is used to compare liver lesions with adjacent liver tissue, the assumption is made that liver tissue is a consistent reference point. The presence of substantial fat, fibrosis, or cirrhosis in the liver, which are common conditions, will alter the scattering characteristics of liver tissue in ways that are not easily predicted. This problem could result in mischaracterization of lesions in an abnormal liver.

Another clear limitation is that the histologic diagnosis of the malignancy or benignity of a liver tumor generally is made at a microscopic cellular level, which is not detectable with this or other US techniques. For example, many epithelial metastases can incite fibrosis, which in some cases can be prominent, as in Figure 7 part D, which depicts the colonic metastasis evaluated. It appears that this fibrosis altered frequency-dependent backscattering to give this metastasis a slightly negative change in CNR, similar to that of a benign hemangioma.

The biggest limitation to clinical application at the present time is that the technique we used in this study is too mechanistically difficult for routine clinical application. Because it took at least several seconds to switch between high and low frequencies, it proved very difficult in many cases to obtain images of a selected liver lesion through exactly the same plane on both the high- and low-frequency images. The color overlays produced by means of image subtraction were created in an effort to show what may be a more desirable way to incorporate information on backscatter frequency dependence clinically.

In principle, it should be possible to acquire data from a region of an organ such as the liver to produce image data corresponding to two frequencies very rapidly, perhaps in one or two transducer passes. With sufficient dedicated computing power within the acquiring US machine, it may be possible to produce color image overlays that indicate the spatial distribution of regions of relatively high and low frequency dependence of backscattering relative to the background, as shown in Figures 1 part F and 2 part C.

Dual-frequency processing may prove a valuable addition to conventional US examinations and may provide improved characterization of lesions in the liver and perhaps other parts of the body as well. It may also be possible to improve lesion detectability with dual-frequency processing, because some lesions difficult to detect at a single frequency may become more evident, such as the simulated lesion poorly seen in Figure 2 part A but more readily appreciated on the dual-frequency image in Figure 2 part C.


   Acknowledgments
 
The authors thank Terry Desser, MD, who reviewed the manuscript and offered helpful suggestions.


   Footnotes
 
Abbreviation: CNR = contrast-to-noise ratio

Author contributions: Guarantor of integrity of entire study, G.S.; study concepts, G.S.; study design, E.W.O., G.S.; definition of intellectual content, E.W.O., G.S.; literature research, G.S.; clinical studies, G.S., E.W.O.; experimental studies, G.S., E.W.O.; data acquisition, G.S., E.W.O.; data analysis, G.S., E.W.O., L.T.; statistical analysis, E.W.O.; manuscript preparation, editing, and review, G.S., E.W.O., L.T.


   References
TOP
 Abstract
 Introduction
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 References
 

  1. Insana MF, Wagner RF, Brown DG, Hall TJ. Describing small-scale structure in random media using pulse-echo ultrasound. J Acoust Soc Am 1990; 87:179-192.[Medline]
  2. McDicken WN. Diagnostic ultrasonics: principles and use of instruments 3rd ed. New York, NY: Wiley, 1991.
  3. Coleman DJ, Lizzi FL. Computerized ultrasonic tissue characterization of ocular tumors. Am J Ophthalmol 1983; 96:165-175.[Medline]
  4. Insana MF, Hall TJ, Fishback JL. Identifying acoustic scattering sources in normal renal parenchyma from the anisotropy in acoustic properties. Ultrasound Med Biol 1991; 17:613-626.[Medline]
  5. Insana MF. Ultrasonic imaging of microscopic structures in living organs. Int Rev Exp Pathol 1996; 36:73-92.[Medline]
  6. King DL, Lizzi FL, Feleppa EJ, et al. Focal and diffuse liver disease studies by quantitative microstructural sonography. Radiology 1985; 155:457-462.[Abstract/Free Full Text]
  7. Yamashita Y, Ogata I, Urata J, Takahashi M. Cavernous hemangioma of the liver: pathologic correlation with dynamic CT findings. Radiology 1997; 203:121-125.[Abstract/Free Full Text]
  8. Tung GA, Vaccaro JP, Cronan JJ, Rogg JM. Cavernous hemangioma of the liver: pathologic correlation with high-field MR imaging. AJR 1994; 162:1113-1117.[Abstract/Free Full Text]
  9. Hanafusa K, Ohashi I, Himeno Y, Suzuki S, Shibuya H. Hepatic hemangioma: findings with two-phase CT. Radiology 1995; 196:465-469.[Abstract/Free Full Text]
  10. Rosai J. Ackerman's surgical pathology 8th ed. St Louis, Mo: Mosby–Year Book, 1996.
  11. Leslie DG, Johnson CD, Johnson CM, Ilstrup DM, Harmsen WS. Distinction between cavernous hemangiomas of the liver and hepatic metastases on CT: value of contrast enhancement patterns. AJR 1995; 164:625-629.[Abstract/Free Full Text]
  12. Hillman BJ, Smith EH, Gammelgaard J, Holm HH. Ultrasonographic-pathologic correlation of malignant hepatic masses. Gastrointest Radiol 1979; 4:361-365.[Medline]
  13. Stetson PF, Sommer FG, Macovski A. Lesion contrast enhancement in medical ultrasound imaging. IEEE Trans Med Imaging 1997; 16:416-425.[Medline]
  14. Sommer FG, Stetson P, Chen HS, Stern RA, Rachlin DJ, Macovski A. Prospects for ultrasonic spectroscopy and spectral imaging of abdominal tissues. J Ultrasound Med 1993; 12:83-90.[Abstract]
  15. Lizzi FL, Astor M, Feleppa EJ, Shao M, Kalisz A. Statistical framework for ultrasonic spectral parameter imaging. Ultrasound Med Biol 1997; 23:1371-1382.[Medline]
  16. Stetson P, Sommer G. Ultrasonic characterization of tissues via backscatter frequency dependence. Ultrasound Med Biol 1997; 23:989-996.[Medline]
  17. Huisman HJ, Thijssen JM, Wagener DJ, Rosenbusch GJ. Quantitative ultrasonic analysis of liver metastases. Ultrasound Med Biol 1998; 24:67-77.[Medline]



This article has been cited by other articles:


Home page
 J Ultrasound MedHome page
T. Wilson, Q. Chen, J. A. Zagzebski, T. Varghese, and L. VanMiddlesworth
Initial clinical experience imaging scatterer size and strain in thyroid nodules.
J. Ultrasound Med., August 1, 2006; 25(8): 1021 - 1029.
[Abstract] [Full Text] [PDF]

Home page
 J Ultrasound MedHome page
K. T. Szopinski, M. Wysocki, A. M. Pajk, R. Z. Slapa, W. Jakubowski, and M. Szopinska
Tissue Harmonic Imaging of Thyroid Nodules: Initial Experience
J. Ultrasound Med., January 1, 2003; 22(1): 5 - 12.
[Abstract] [Full Text] [PDF]

Home page
 J Ultrasound MedHome page
T. Wilson, J. Zagzebski, and Y. Li
A Test Phantom for Estimating Changes in the Effective Frequency of an Ultrasonic Scanner
J. Ultrasound Med., September 1, 2002; 21(9): 937 - 945.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Sommer, G.
Right arrow Articles by Tai, L.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Sommer, G.
Right arrow Articles by Tai, L.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE