| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Ultrasonography |
1 From the Imaging Sciences Department, Hammersmith Hospital, Imperial College, 150 Du Cane Rd, London W12 0HS, England (T.H.B., M.J.B., C.J.H., M.L., D.O.C.). The complete list of author affiliations and the author contributions are cited at the end of this article. From the 2002 RSNA scientific assembly. Received April 15, 2003; revision requested July 2; final revision received January 15, 2004; accepted March 8. Supported by Siemens Medical Solutions, Ultrasound Division. Address correspondence to T.H.B. (e-mail: thomas.bryant@ic.ac.uk).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
MATERIALS AND METHODS: One hundred forty-two patients with liver lesions underwent conventional gray-scale and color Doppler US and SH U 508A–enhanced US. Two radiologists blindly read digital cine clips and assigned scores for confidence in diagnosis of benignancy or malignancy, diagnosis of specific lesion types, and relative difference in SH U 508A uptake between the lesion and the liver parenchyma (ie, subjective conspicuity score [SCS]). Comparisons were made to see whether the addition of agent detection imaging led to improved diagnostic performance.
RESULTS: Receiver operating characteristic analysis revealed improved discrimination of benign and malignant lesions for readers 1 (P = .049) and 2 (P < .001). The number of patients with a correct diagnosis of benignancy or malignancy assigned by readers 1 and 2, respectively, improved from 114 and 113 to 125 and 128 with agent detection imaging (reader 1: P = .027; reader 2: P = .008; McNemar test). Specific diagnoses were made more accurately with agent detection imaging: At McNemar testing, the number of correct lesion type determinations increased from 83 to 92 (P = .022) for reader 1 and from 85 to 99 (P < .001) for reader 2. Both readers assigned high scores for differences in SH U 508A uptake between the liver parenchyma and the lesion for metastases and cholangiocarcinomas and low scores for uptake differences in most of the benign lesions. Hepatocellular carcinomas (HCCs), hemangiomas, and adenomas had more variable uptake differences. Fourteen of 22 hemangiomas were assigned an SCS of less than 50%, and 22 (reader 1) and 15 (reader 2) of 31 HCCs were assigned an SCS of greater than 50%.
CONCLUSION: With use of SH U 508A–enhanced agent detection imaging, liver lesion characterization and diagnostic performance are significantly improved.
© RSNA, 2004
Index terms: Liver neoplasms, 761.3192, 761.3194, 761.3198, 761.321, 761.323, 761.33 • Liver neoplasms, diagnosis, 761.30 • Liver neoplasms, US, 761.12983, 761.12984, 761.12988, 761.12989 • Microbubbles • Ultrasound (US), contrast media, 761.12988
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
The results of several studies have shown the usefulness of the effect of this late hepatosplenic phase in assessing focal lesions. At imaging, many malignant lesions manifest as enhancement abnormalities in the normal liver parenchyma, whereas most benign lesions show contrast material uptake. Malignancies normally show very little uptake because they lack normal liver tissue and do not accumulate microbubbles. Hemangiomas and focal nodular hyperplasia (FNH) show substantially higher uptake, and focal fatty sparings or changes and regenerating nodules usually show marked uptake owing to their similarity to the adjacent liver parenchyma (4–9).
In most of the related previous studies, pulse-inversion US modes (6–9) were used, but the paired acquisition of registered conventional gray-scale US and microbubble US scans would have obvious advantages when there was microbubble uptake in lesions. The findings in a single-center study involving the use of a color Doppler US method that was not optimized for microbubble-enhanced imaging but allowed paired gray-scale US scan and microbubble US scan registration (4,5) showed excellent discrimination between benign and malignant lesions. Recently, agent detection imaging (Siemens Medical Solutions, Ultrasound Division, Mountain View, Calif) has become available. Agent detection imaging involves the use of a dedicated paired-image capture microbubble mode that is designed to have good space-filling properties and higher spatial resolution than color Doppler US (4,5).
Our purpose in the current study was to evaluate in a prospective multicenter study whether the characterization of focal liver lesions can be improved by using SH U 508A–enhanced agent detection imaging, as compared with the lesion characterization achieved with nonenhanced B-mode and color or power Doppler US.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
This prospective study was conducted at four centers (Hammersmith Hospital, London, England; Benjamin Franklin University Hospital, Berlin, Germany; Kings College Hospital, London, England; and Glasgow Royal Infirmary, Glasgow, Scotland) in Britain and Germany between January 2001 and February 2002. All patients gave written informed consent, and the studies were approved by the local research ethics committees at all of the participating centers.
US Imaging
For consistency in imaging technique and analysis, all US examinations were performed by using the same system (Sequoia 512 with a 4C1, 4V1, or 6C2 probe; Siemens Medical Solutions, Ultrasound Division). All scanning was performed by experienced radiologists or by a technician supervised by a radiologist. The involved radiologists were M.J.B., T.A., P.S.S., L.H.B., C.W.H., C.J.H., and D.O.C. The technician was M.L., and the research assistant was J.M. The levels of US experience and microbubble US experience of these individuals were, respectively, 13 and 8 years (M.J.B.), 10 and 6 years (T.A.), 13 and 7 years (P.S.S.), 4 and 6 months (L.H.B.), 7 and 3 years (C.W.H.), 10 and 4 years (C.J.H.), 25 and 12 years (D.O.C.), 12 and 5 years (M.L.), and 3 and 3 years (J.M.). To standardize the US imaging technique used at the participating centers, a study protocol form was created by one of the investigators (M.J.B.) and then approved by each of the participating centers.
Initially, the liver was scanned by using gray-scale US and a target lesion was identified. In this study, only one lesion per patient was scanned. When more than one lesion was seen, the supervising radiologist chose the lesion with the best possible acoustic window. Baseline digital cine clips that were a few seconds long were obtained. Baseline US scanning was performed by using both harmonic and fundamental gray-scale US examinations with system default abdominal imaging settings chosen to optimally image the lesion according to the opinion of the supervising radiologist. After gray-scale US was performed, color Doppler US or power Doppler US was performed to again optimally visualize the lesion according to the opinion of the supervising radiologist. The sonologists were permitted to freely vary the system settings to obtain optimal scans. The sonologists were not blinded to the final diagnosis if it was available at the time of the study.
In each patient, one 2.5-g bolus of SH U 508A was injected at a concentration of 300 mg/mL intravenously and followed by a flush of 10 mL of normal saline. The bolus was repeated with a higher dose, 4.0 g, if it was judged by the local investigator to be insufficient to yield a diagnostic scan. To avoid destruction of the microbubbles during the blood pool phase, no scanning was permitted for 5 minutes. After this interval, the lesion was localized by using gray-scale US. Then, agent detection imaging was performed by using a predefined region of interest (defined by a color box) of a size and at a site that were sufficiently large to enclose the lesion. The agent detection imaging mode involved the use of a Doppler frequency of 2.5–4.0 MHz, with maximum acoustic power (mechanical index = 1.8–1.9) and pulse-repetition frequencies.
After an analysis of images acquired early in the study, a subset of 59 patients received a second injection of SH U 508A, and after 2.5 minutes an additional US scan (ie, 2.5-minute scan) was obtained and analyzed separately to see if imaging at this time yielded better information. No adverse affects were noted secondary to the contrast agent administrations.
All US scans used for analysis were stored as short digital video clips. Backup S-VHS videotaped images were recorded, and in some instances, these tapes were redigitized into digital cine clips for later analysis if the digital storage process failed. The clips were collected for central-location review by a radiologist (T.H.B.) who was blinded to the reference-standard diagnoses. The clips were then screened for technical quality.
Reference-Standard Diagnosis
Before the start of the study, the lesion characterization criteria that would be used to establish the reference-standard diagnoses were agreed on at an investigators meeting attended by the senior radiologist from each site. These criteria are listed in the Appendix. The biopsy result was accepted as the reference diagnosis for all lesion types. If biopsy was not performed, a lesion could be included if it met the defined multimodality imaging and/or follow-up criteria. The only change to the protocol as originally agreed on between the investigators was that adenomas were initially omitted from the initial list of lesions and added later.
Blinded Readings
The digital cine clips were read by two independent radiologists from institutions not involved in scanning patients: R.B., who had 11 years of conventional US experience and 9 years of contrast-enhanced US experience, and J.M.P., who had 8 years of conventional US experience and 3 years of contrast-enhanced US experience. These readers were told only the age and sex of each patient and whether the patient had received a diagnosis of cirrhosis. No other clinical information was given, and the patients were identified only by an alphanumeric code.
All readings were performed on the same computer (Macintosh Titanium Powerbook with 15.2-inch TFT display; Apple, Cupertino, Calif) at a central location. Each reader was presented with digital cine clips in the following order: gray-scale US scans, color or power Doppler US scans, and agent detection scans obtained 5 minutes after SH U 508A administration (gray-scale only clip to determine lesion location and the combined, paired gray-scale and agent detection imaging color information). For each case, a single frame from the gray-scale cine US scan file was shown with the lesion circled (by T.H.B. or M.J.B.) to clarify any ambiguity regarding the position of the lesion at analysis. The readers were allowed to adjust the ambient lighting to an optimal level and could take as long as they needed to read the images in each case.
For each clip, the readers were asked to grade the lesion as most likely benign or most likely malignant, rate the probability of malignancy on a visual analog scale—with 0 meaning definitely benign and 100 meaning definitely malignant—and give the most likely etiologic diagnosis. For the agent detection scans, the readers were also asked to score the uptake of SH U 508A in the lesion and in the adjacent liver parenchyma by using a percentage scale. The difference between the background uptake score and the lesion uptake score was calculated as a subjective conspicuity score (SCS).
When an additional 2.5-minute scan was obtained, it was presented for interpretation after the 5.0-minute agent detection scan was presented, and the same criteria (described earlier) that were used to assess lesions seen on the 5.0-minute scans were used.
Statistical Analyses
All data obtained at the blinded readings were recorded on a spreadsheet by a clerical research coordinator who had no other involvement in the study. These data were then analyzed by a radiologist investigator (T.H.B.) who did not perform any scanning and was advised by a professional medical statistician.
Comparisons of the numbers of male and female patients were performed by using the binomial test, and age comparisons between the male and female patient groups were performed by using a t test with equal variances not assumed. The diagnostic performance of pre- and postcontrast US in the assessment of malignancy was compared by using areas under the receiver operating characteristic (ROC) curves (in a binormal model accounting for paired-data covariance) derived from the visual analog scores of malignancy (with two-tailed P values). The sensitivity, specificity, and accuracy of US in the assessment of malignancy and in the diagnosis of specific lesion types before and after contrast material administration were calculated by using the reference diagnosis as the reference standard.
Differences between the pre- and postcontrast lesion scores and between the 5.0- and 2.5-minute agent detection scans were analyzed by using the McNemar test. Comparisons of SCS values among the lesion categories were made by using Dunn multiple-comparison, Kruskal-Wallis analysis of variance. ROC curves were also drawn to evaluate the performance of SCS values alone in determining malignancy. Weighted 
statistics were used to examine degrees of interobserver variation in visual analog scores of malignancy and SCS values.
For the hepatocellular carcinomas (HCCs) for which biopsy was performed, a linear regression analysis of SCS values by histologic grade was performed. Comparisons of SCS values between the 5.0- and 2.5-minute agent detection scans were based on comparisons of diagnostic accuracy by using the McNemar test and are also presented in graphical format for visual comparison.
ROC analysis was performed with dedicated software (ROCKIT; Charles E. Metz, University of Chicago, Chicago, Ill), and apart from Kruskal-Wallis analysis of variance (Unistat for Excel 5.0; London, England) and statistics, which were calculated by using the formula in and verified by using test data from Altman (10) with an Excel macro (Microsoft, Redmond, Wash) written by the investigators, other analyses were performed with standard statistical software (SPSS 11.0; SPSS, Chicago, Ill).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
1. In 13 patients, a reference diagnosis that fulfilled the criteria established at the start of the study could not be made.
2. The cases of 11 patients were excluded from analysis by the supervising radiologist (T.H.B.) because the US scans obtained were technically unsatisfactory: One case involved US system failure. One case was that of a patient whose baseline scans showed a lesion different from that seen on the agent detection scans in the opinion of the supervising radiologist, so a comparison could not be made. Six cases were those of patients in whom the lesion could not be seen on the digital cine clips provided. In three cases, although US scans were available, the gray-scale US images had been obtained with incorrect gain settings and thus were almost completely black.
3. Four patients did not meet the study protocol criteria: One patient had a lesion outside the liver (in the right adrenal gland at biopsy). For two patients, the reference diagnosis of the lesion type (liver abscess) was not on the agreed upon list of eligible lesions. The medical and imaging records of one patient were entered into the database even though the lesion was not visible on the initially obtained US scans.
4. There was one case where a technical failure occurred because attenuation of the ultrasound beam in the liver prevented the acquisition of a satisfactory agent detection scan. In this patient, the lesion was 13 cm deep from the skin surface and the agent detection imaging effect was reduced beyond 12 cm (although contrast was seen more superficially).
5. For five patients, no US scans were available: For four patients, no images were forwarded to the central center for analysis, and for one patient, the US examination was stopped before agent detection scans were obtained.
Histopathologic information was available for 63 (44%) of the 142 examined lesions: 52 (59%) of 88 malignant lesions and 11 (20%) of 54 benign lesions. The malignant lesions consisted of 50 metastases, 31 HCCs, and seven cholangiocarcinomas. The benign lesions consisted of 22 hemangiomas and 32 benign nonhemangiomatous lesions: four adenomas, four cases of focal fatty change, five cases of focal fatty sparing, two regenerating nodules, and 17 cases of FNH. Twenty-three patients had cirrhosis, but only 18 of the 31 patients with HCC had cirrhosis. Examples of the different cases are shown in Figures 1 and 2.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
The 50 metastases originated from 23 colorectal adenocarcinomas, two esophageal carcinomas, three breast adenocarcinomas, two pancreatic adenocarcinomas, one lung adenocarcinoma, four adenocarcinomas from unknown primary sites, four neuroendocrine tumors, three melanomas, two non-Hodgkin lymphomas, one gastrointestinal stromal tumor, one testicular carcinoma, one transitional cell carcinoma, and three undifferentiated carcinomas from unknown primary sites.
Results are presented in terms of the improvement gained by using agent detection imaging, as compared with baseline US scanning, in the differentiation between benign and malignant lesions, the accurate diagnosis of specific lesion types, and interobserver agreement (in statistics), and in terms of SH U 508A uptake differences (in SCS values) between the liver parenchyma and the lesion and 2.5-minute US scan findings.
Differentiation between Benign and Malignant Lesions
The results of ROC analysis are shown in Figure 3. For both readers, there were significant increases in areas under the ROC curves with agent detection imaging: from 0.88 at baseline US to 0.94 at agent detection imaging (P = .049) for reader 1 and from 0.88 at baseline US to 0.98 at agent detection imaging (P < .001) for reader 2. The data in Table 1 show that in 142 patients, the classification of lesions as benign or malignant was more often correct with agent detection imaging than with conventional US (P = .027 for reader 1, P = .008 for reader 2).
|
|
|
|
SCS Values
SCS distributions for each type of lesion are shown in Figure 4. The results of comparisons of SCS values between lesion types are shown in Table 3, and the assigned SCS values for each lesion type are shown in Table 4. For both readers, there were significant differences in agent detection imaging SCS values between the benign nonhemangiomatous lesion group and all three malignant lesion groups (P = .001 for HCCs, P < .001 for cholangiocarcinomas, and P < .001 for metastases). Hemangioma SCS values were significantly different from metastasis SCS values (P < .001). However, only reader 1 assigned significantly different SCS values between hemangiomas and cholangiocarcinomas (P = .037 for reader 1, P = .17 for reader 2), and neither reader assigned significantly different SCS values between hemangiomas and HCCs (P = .32 for reader 1, P > .99 for reader 2). Differences in SCS values were mainly due to differences in lesion uptake scores rather than differences in background liver uptake scores. Median liver background scores were 100% (range, 50%–100%) for reader 1 and 93% (range, 30%–100%) for reader 2.
|
|
|
|
|
Benign nonhemangiomatous lesions generally were judged to have high uptake. Areas of focal fatty change and sparing showed very high uptake: An SCS of 0% was assigned by reader 1 in eight of nine cases and by reader 2 in six of nine cases (median SCS for both readers, 0%). FNH also showed high uptake (median SCS, 0%); all scores assigned by both readers, except those of two patients for each reader, were less than 15%. Only two proved regenerating nodules were included, and the two readers scored these lesions differently: One reader assigned scores of 0% and 60%, and the other assigned scores of 5% and –5%. Adenomas had the most variable SCS values in the benign nonhemangiomatous lesion group; scores ranged from 0% to 100%.
Hemangiomas showed great variability in uptake, with median SCS values of 40% for reader 1 and 31% for reader 2. However, both readers assigned 14 of 22 hemangiomas an SCS of less than 50%, and only six hemangiomas were diagnosed as malignant by reader 1 and only three were diagnosed as malignant by reader 2.
SCS values were also plotted on ROC curves and showed diagnostic capability in the determination of malignancy at comparison of scores between conventional US alone and postcontrast agent detection imaging (area under ROC curve, 0.92 for reader 1 and 0.87 for reader 2).
Statistics
Interobserver agreement between the two readers improved with use of agent detection imaging: The weighted of 0.44 for the visual analog scale score for diagnostic confidence in determining malignancy and benignancy before SH U 508A injection improved to 0.62 after SH U 508A administration. There was good agreement regarding assigned SCS values between the two readers; the weighted was 0.62.
2.5-Minute US Scans
US scans were obtained 2.5 minutes after SH U 508A administration in 59 patients, who had 18 metastases, two cholangiocarcinomas, 18 HCCs, 11 benign nonhemangiomatous lesions, and 10 hemangiomas. The data illustrated in Figure 5 demonstrate that there was little difference in uptake between the scans obtained 2.5 minutes and those obtained 5.0 minutes after contrast agent administration. There was no significant improvement in diagnostic accuracy: Reader 1 correctly diagnosed 51 (86%) of the 59 lesions as benign or malignant at 2.5 minutes and 53 (90%) lesions as benign or malignant at 5.0 minutes (P = .5). Reader 2 diagnosed 48 (81%) of the 59 lesions as benign or malignant at 2.5 minutes and 49 (83%) lesions as benign or malignant at 5.0 minutes (P > .99).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
This technique does not require a separate person to perform the injection. In addition, it appears that the time of image acquisition is not critical: We did not observe a substantial difference between scans acquired 2.5 minutes and those acquired 5.0 minutes after contrast agent administration. Agent detection imaging enables one to characterize the benignancy or malignancy of lesions with high accuracy and aids in the diagnosis of common liver lesion types with a high level of accuracy.
We attempted to make the study protocol as similar to actual clinical practice as possible. The use of short video clips allows some of the real-time nature of US to be incorporated into blinded readings. The ages and sexes of the patients and the presence of cirrhosis in the study population mimicked those in a typical minimum clinical data set; however, the information about cirrhosis may have been actively misleading because only 23 of the total of 142 patients and only 18 of the 31 patients with HCC were known to have cirrhosis.
The order of the blinded readings matched that in a normal clinical US practice in that we assessed the effect that adding agent detection data readings to initial readings of gray-scale and color Doppler data has on diagnostic performance. Contrast-enhanced US scanning would not be performed without preliminary baseline scanning having been performed first, and the approach of comparing baseline nonenhanced US scans with postcontrast US scans has been used previously (11).
To our knowledge, this is the first multicenter study with results that show the diagnostic improvement achieved with agent detection imaging in clinical practice. We are unaware of any other similar multicenter studies. In a recent single-center study, pulse-inversion US scanning performed 2.5 minutes after SH U 508A administration (7) had an accuracy of 85% in the differentiation between benign and malignant lesions, with an area under the ROC curve of 0.947; these data are very similar to the results observed in our current study (accuracies of 88% and 90% for the two readers, with corresponding ROC areas of 0.94 and 0.98, respectively).
The uptake of liver-specific agents (eg, superparamagnetic iron oxide [SPIO] particles, manganese dipyridoxyl diphosphate [Mn-DPDP], and gadobenate dimeglumine) by benign lesions at magnetic resonance (MR) imaging has been described (12–17). Malignant lesions usually show little uptake, although HCCs can be an exception (18). In multicenter (15) and single-center (17) trials of lesion differentiation with computed tomography (CT) and MR imaging, the accuracies achieved and specific diagnoses made were similar to those in our study. CT, Mn-DPDP–enhanced MR imaging, and gadobenate dimeglumine–enhanced MR imaging had accuracies of 68%, 85%, and 91%, respectively (17), in the differentiation of malignant and benign lesions, similar to the accuracies of 88% and 90% for the two readers in our study. Contrast-enhanced US, however, is less expensive than contrast-enhanced MR imaging, and agent detection imaging can be easily performed immediately after a lesion is discovered and after gaining intravenous access. Thus, agent detection imaging enables one to avoid the time delay and patient anxiety involved in scheduling another examination.
The use of SCS values removes much of the "skill" element from the interpretation of postcontrast US scans. SCS values are based simply on the difference in uptake between the lesion and the background liver. ROC analysis of SCS values revealed that the consideration of this single factor alone can yield data that are at least as accurate in the classification of a lesion as benign or malignant as those obtained by our US specialists when they used only gray-scale US and color Doppler US.
HCCs had very variable uptake and were difficult to differentiate from benign lesions—particularly hemangiomas—on the basis of SCS values alone. There are many reports of hemangiomas showing peripheral globular or rimlike enhancement with progressive fill-in at dynamic contrast-enhanced US scanning, a finding that is often persistent into the late phase (6,8,9,19). HCCs can be more heterogeneous in appearance, both in cirrhotic and noncirrhotic livers (8,20,21). The readers in our study had greater accuracy in distinguishing hemangiomas from HCCs than is possible with use of SCS values alone. These results could be interpreted as suggesting that the pattern of contrast enhancement was also important in the diagnosis. Although the readers were not specifically asked about patterns of enhancement, we noted that hemangiomas often showed a peripheral enhancement pattern at agent detection imaging (Fig 2h). In addition, hemangiomas are seen less commonly in patients with cirrhotic livers than in the general population (22,23); this characteristic may aid in the diagnosis.
Previous study results (13,24,25) have shown that the grade of HCC is closely related to the uptake of SPIO particles and Mn-DPDP, owing to the presence of Kupffer cells in the lesion in the case of SPIO particle uptake and to the presence of hepatocytes in the lesion in the case of Mn-DPDP uptake. Well-differentiated lesions can show generally homogeneous enhancement, whereas poorly differentiated lesions can show irregular patchy enhancement or heterogeneous enhancement. A positive correlation between tumor grade and SH U 80A uptake has been previously reported (26). However, these findings were not reproduced with SH U 80A in the current study, and there was only a very weak correlation between lesion grade and SCS.
There were important limitations to our study. First, not all lesions were sampled at biopsy; however, the diagnoses were determined by using a rigorous technique of multimodality imaging and long-term follow-up in combination with other clinical information. The absence of biopsy results in some cases introduced the potential for errors in diagnosis, particularly between the different types of benign lesions or between the subgroups of malignant lesions. In particular, the differentiation of adenomas, FNH, and regenerating nodules could have been affected because the imaging characteristics of these lesions are similar at CT and MR imaging. It is reassuring, however, that these are all benign lesions. We believe that it would not have been ethical to perform biopsy in patients with lesions that would be considered benign on the basis of all available imaging data or in whom the results of biopsy of a metastasis or HCC would have made no difference in the treatment or clinical outcome.
Second, we studied only late-phase US of liver lesions with the microbubble contrast agent. It has been previously shown that examination of the vascular (ie, blood pool) phase improves diagnostic accuracy in the characterization of focal liver lesions (6–9). It is likely that a combination of vascular and late-phase imaging examinations would lead to further improvements in the diagnostic accuracy of the contrast-enhanced imaging examination. Newer US contrast agents allow real-time scanning at low mechanical index, which causes minimal microbubble disruption throughout the vascular phase and into the late phase. However, the approach that we describe herein is technically simple and fast and requires only one operator.
Third, our subjective grading of lesion conspicuity had limitations. There was some interobserver variation, although there was good agreement regarding SCS values; the weighted was 0.62. A single SCS for uptake by the lesion was required in the protocol, but some lesions show variation in uptake—this is particularly prevalent in hemangiomas and HCCs—and thus can be difficult to score. The readers were asked to score the area of highest uptake, however. This may explain some of the variation between the scores assigned by the two readers. It is noteworthy that the blinded readers did well at characterizing both HCCs and hemangiomas, despite variability in SCS values.
Fourth, this study was performed at centers with operators who had considerable experience in the use of contrast-enhanced US, and, thus, the results may not reflect those that would be obtained in centers with less experienced operators. However, late-phase US is a simple technique that can be performed by sonographers with less experience in the use of microbubble contrast agents.
Fifth, there were potential sources of bias. We recruited only patients with lesions larger than 10 mm in diameter, and larger lesions can be more easily characterized. In addition to being transient—in that the microbubbles are destroyed during the acquisition of the images, so the image appears on screen for only a fraction of a second—agent detection imaging is limited in depth and focal zone, although it is less limited in these features compared with earlier microbubble US modes relying on the "loss of correlation" technique. The effect of agent detection imaging is reduced at depths greater than approximately 12 cm and in the presence of fatty hepatic change. Sharply defined zones of absent effect are seen when previous insonation has occurred. Furthermore, the limitations of conventional US in terms of patient body habitus also apply to agent detection imaging. Not all lesions can be analyzed, and not all patients are suitable for this technique. There were a few technical failures in this study, and there may have been selection bias in favor of the more easily visualized lesions.
Sixth, digital cine clips were presented to the readers; conventional US scans were displayed first, and contrast-enhanced agent detection scans were displayed next. This order could have introduced bias because more views of each lesion were seen as the reading progressed. However, this could not be avoided because we were attempting to mimic the order of examinations and readings performed in the clinical use of microbubble-enhanced US.
In conclusion, agent detection imaging performed during the late phase of SH U 508A uptake is superior to conventional US in the differentiation between benign and malignant liver lesions and in the diagnosis of specific lesion types. Malignant lesions typically appear as uptake abnormalities, whereas benign lesions show prolonged uptake of the microbubble contrast. Agent detection imaging is a rapid technique that can help improve diagnostic ability and aid in reducing the need for further investigations. Owing to the wide availability of SH U 508A and US, the use of agent detection imaging can influence clinical practice in many countries.
|
APPENDIX |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXREFERENCES |
|---|
HCC: Biopsy results showing HCC or a hypervascular nodule seen at angiography, CT, or MR imaging in a patient with biopsy-proved cirrhosis and progressive enlargement of the lesion at serial imaging.
Metastasis or lymphoma: Biopsy results showing metastasis or lymphoma or a biopsy-proved malignancy at another, nonhepatic site, with characteristic liver metastasis at CT or MR imaging that had been shown to be progressively enlarging (or reducing in size with chemotherapy) at sequential cross-sectional imaging.
Cholangiocarcinoma: Only biopsy or cytologic confirmation was acceptable.
Hemangioma: Biopsy results showing hemangioma or dynamic contrast-enhanced CT or MR imaging findings reported as showing diagnostic features of hemangioma with no change at follow-up of longer than 1 year.
Regenerating nodules: Biopsy results showing regenerating nodules, normal 
-fetoprotein levels combined with either normal liver-specific contrast agent (ie, SPIO or Mn-DPDP)–enhanced MR imaging findings, or sequential CT, MR, or US examinations revealing no progression for more than a year.
FNH: Biopsy results showing FNH, late-phase uptake documented at either SPIO-enhanced MR imaging or liver colloid scanning
