| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Gastrointestinal Imaging |
1 From the Mallinckrodt Institute of Radiology (S.A.T., C.C.H., F.D., B.A.S., J.P.H., J.J.B., E.G.M., W.D.M., D.M.B.), Departments of Internal Medicine (M.G.P.) and Pathology (J.H.R.), and Siteman Cancer Center (S.A.T., C.C.H., F.D., B.A.S., J.P.H., J.J.B., E.G.M., W.D.M., D.M.B., J.H.R.), Washington University School of Medicine, 510 S Kingshighway Blvd, St Louis, MO 63110. Received December 4, 2001; revision requested February 18, 2002; revision received April 8; accepted May 23. Address correspondence to S.A.T. (e-mail: teefeys@mir.wustl.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Twenty-five patients were examined prospectively with CT, MR imaging, US, and PET. Each test result was interpreted independently by two radiologists. Explanted liver specimens were examined histologically to determine presence and type of lesion. Results were analyzed on a patient-by-patient basis with marginal homogeneity and effect likelihood ratio tests.
RESULTS: HCC was diagnosed in nine patients. US diagnostic performance was superior to that of CT and MR imaging on a patient-by-patient basis. Sensitivities were higher for US (0.89 for both US readers) than they were for CT (0.67 and 0.56 for readers 1 and 2, respectively), MR imaging (0.56 and 0.50 for readers 1 and 2, respectively), and PET (0 for both readers). None of the differences (within test) between readers were significant (P 
.32). Ratings by US and MR observers and one CT observer were significantly associated with truth (P 
.04). One or more imaging tests depicted 68 lesions. Histologic analysis revealed 18 HCC nodules; of these, 13 were correctly identified at CT, 14 at MR imaging, 13 at US, and none at PET. There were nine false-positive diagnoses of HCC with CT, five with MR imaging, and nine with US.
CONCLUSION: Although US had the best diagnostic performance in depicting HCC on a patient-by-patient basis and was substantially better than were MR imaging and CT (which had nearly equivalent diagnostic performances), CT, US, and MR imaging performed similarly on a lesion-by-lesion basis. Small tumor nodules were the most common cause of missed HCCs with all tests. PET did not depict any HCCs.
© RSNA, 2003
Index terms: Computed tomography (CT), comparative studies, 761.12114 • Liver neoplasms, 761.323 • Magnetic resonance (MR), comparative studies, 761.121411, 761.121412, 761.121416, 761.12143 • Positron emission tomography (PET), comparative studies, 761.12163 • Ultrasound (US), comparative studies, 761.12981, 761.12983
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
During the past decade, authors of several studies have reported about the accuracy of individual tests or combinations of tests in the diagnosis of HCC (3–15). While authors of earlier articles suggested that tests such as computed tomography (CT) and magnetic resonance (MR) imaging were promising tools in the detection of HCC (5,12), later study findings showed poorer results, particularly, when large surveillance populations were studied (3). Authors of several studies, whose findings were published in the late 1980s to mid 1990s, compared the accuracy of a combination of tests, including CT and ultrasonography (US), in the detection of HCC (9–11,14) and reported varying results. All studies were retrospective, and imaging findings were not compared with the liver explantation results. In these studies, CT was performed without dual phase imaging, with a slow or variable contrast material injection rate, and with a 10-mm section thickness; US was performed with older equipment. Authors of two later studies (13,15), one of which was prospective (13), compared MR imaging with CT in the detection of HCC. In both studies, hepatic arterial, delayed, or dual phase imaging was used; however, in neither study were the imaging findings compared with the pathologic results of the liver explantation. Authors of one study (13) reported that arterial phase dynamic MR imaging was superior to helical CT in the detection of HCC; whereas, authors of the other study (15) concluded that both dynamic MR imaging and early-phase CT were the most sensitive means of detecting HCC.
Recently, positron emission tomography (PET) with fluorine 18 fluorodeoxyglucose (FDG) has begun to play an increasingly important role in helping to detect and stage certain cancers. A few study findings have shown that FDG PET is moderately sensitive in the detection of CCA (16–18) and moderate-to-high grade HCC (19,20). However, no known study findings have been published in which PET was compared with other imaging tests in the detection of HCC or CCA. Additionally, the utility of FDG PET in the detection of primary hepatic tumors in patients who are awaiting transplantation and who have poor hepatic function has not been studied.
The purposes of this prospective study were to (a) determine and compare the diagnostic performance of CT, MR imaging, US, and PET against the standard of histologic examination of the resected liver specimen to assess which single test or combination of tests is most accurate in the detection of HCC or CCA in listed liver transplant candidates and to (b) determine the interobserver variability between independent readers of one test (CT, MR imaging, US, or PET).
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
-fetoprotein level (>30 µg/L) or with primary sclerosing cholangitis were eligible. All patients underwent CT, MR imaging, US, and PET. Repeat CT or MR imaging was performed at a frequency that was based on the results of the original study findings. Patients with lesions suggestive of HCC not amenable to biopsy (based on the size or the location of the lesion or the presence of an uncorrectable coagulopathy) or with lesions having discordant CT and MR imaging findings underwent repeat CT every 6 months; those with negative findings, every year. Only one patient did not undergo a repeat study, as requested. That patient underwent liver transplantation 15 months after the last CT examination.
Ten of the patients either died prior to liver transplantation (without autopsy or biopsy being performed) or their names were removed from the transplant list. Two patients whose names had been on the transplant list for more than 2 years were not included in the study because of an inability to obtain follow-up images. The remaining 25 patients form the study population.
Imaging Techniques
All CT studies (Somatom Plus 4; Siemens Medical Systems, Iselin, NJ) were performed by using a scanning time of 0.75 second, collimation of 5 mm, and table feed of 8 mm per gantry rotation. Images were reconstructed at 5-mm intervals. Prior to the administration of contrast material, images were obtained through the entire liver with a 10-mm collimation and a table feed of 10 mm per gantry rotation. Patients received 150 mL of contrast material (ioversol [Optiray 350]; Mallinckrodt, St Louis, Mo) that was injected intravenously at a rate of 5 mL/sec. Images were obtained during the hepatic arterial phase (25 seconds after the start of injection) and the portal venous phase (55 seconds after the start of injection). Delayed images were obtained through the entire liver at 5 minutes in patients at risk of having HCC and at 15 minutes in patients at risk of having CCA.
All MR imaging studies were performed with a 1.5-T MR system (Vision or Symphony; Siemens Medical Systems, Iselin) with a phased-array coil. A section thickness of 8 mm with a 2-mm gap was used during all sequences. Images were initially acquired with conventional transverse T1-weighted spin-echo, T2-weighted fast spin-echo, and in-phase and opposed-phase two-dimensional gradient-echo sequences. Gadodiamide (Omniscan; Nycomed-Amersham, Princeton, NJ) was then injected intravenously at doses of 0.1 mmol per kilogram of body weight at a rate of 2 mL/sec, and transverse two-dimensional gradient-echo images were obtained during the hepatic arterial (20 seconds after the start of the injection) and portal venous (50–60 seconds after the start of the injection) phases. Delayed images were obtained at the same time intervals as were CT scans.
All US studies (HDI 3000, Advanced Technology Laboratories, Bothell, Wash; or Elegra, Siemens Medical Systems, Issaquah, Wash) were performed with a variety of transducers (range: 2–7 MHz). Real-time scanning of the entire liver was performed in transverse and longitudinal planes in a standardized fashion and videotaped by one observer (W.D.M.). Low flow settings for color Doppler imaging were used to assess the presence of internal flow in all detected lesions.
All FDG PET studies were performed with a scanner (Ecat Exact; Siemens/CTI, Knoxville, Tenn) that allows simultaneous collection of 47 transverse sections spanning 16.2 cm. Patients fasted for at least 4 hours prior to scanning. To ensure a fasting state, a blood sample was obtained prior to FDG administration to determine the blood glucose level (none of the patients had fasting hyperglycemia). FDG (555 MBq) was injected intravenously, and images were obtained with the patient in the supine position beginning approximately 60 minutes after injection. Two overlapping 10–15-minute emission scans were obtained centered at the liver. After each emission scan, a 2-minute transmission scan was obtained with a rotating germanium 68 or gallium 68 rod source. To minimize image artifacts due to renal pelvicalyceal activity, patients were hydrated intravenously, 20 mg of furosemide (American Regent, Shirley, NY) was administered, and a urinary catheter was placed as previously described (21). PET images were reconstructed by using a filtered back projection with a Hann filter (frequency cutoff, 0.6 x Nyquist frequency). Emission images were corrected for attenuation with use of the segmentation method (22). Images were displayed in three orthogonal projections and as whole-body maximum pixel reprojection images for visual interpretation.
Image Interpretation
Results of each imaging test (CT, MR imaging, US, and PET) were interpreted independently by two radiologists experienced (>10 years of experience for all radiologists) with the test, for a total of eight observers (S.A.T., W.D.M., F.D., B.A.S., J.P.H., D.M.B., J.J.B., E.G.M.). All readers were blinded to the results of the other tests, as well as to each other’s interpretation. However, readers were not blinded to the patient diagnosis of either primary sclerosing cholangitis or hepatitis. A video-recorded tape was reviewed by the second radiologist interpreting the US study for purposes of data collection. All imaging tests were interpreted for the presence of a hepatic tumor (HCC or CCA). Each reviewer specified the total number of lesions detected (as many as six) per patient and marked their location on a representative diagram that divided the liver into segments and seven transverse sections. The size of each visualized lesion was determined in the transverse plane (anteroposterior and transverse) on all tests. For each patient, the reviewer was asked to indicate his or her degree of confidence that a malignancy was present on the basis of a six-point confidence scale: 1, definitely present; 2, probably present; 3, possibly present; 4, possibly not present; 5, probably not present; and 6, definitely absent.
CT criteria.—The following criteria were used to evaluate HCC. A well-defined, heterogeneously enhancing, and hyperattenuating lesion on arterial phase images or a lesion associated with vascular invasion was rated 1. A well-defined, homogeneously enhancing, and hyperattenuating area on arterial phase images was rated 2. An ill-defined hyperattenuating area on arterial phase images, a heterogeneously enhancing and hypoattenuating lesion on portal venous phase images, or a homogeneous hypoattenuating lesion on portal venous phase images was rated 3 or 4. A well-defined lesion measuring near water attenuation value (features typical of a simple cyst) was rated 5 or 6.
The following criteria were used to evaluate CCA. A lesion showing mild peripheral enhancement on arterial and portal venous phase images, with pooling of contrast material on delayed images, and being associated with bile duct dilatation peripheral to the lesion was rated 1. A hypoattenuating lesion on portal venous phase images associated with bile duct dilatation peripheral to the lesion, without pooling of contrast material on delayed images, was rated 2. A hypoattenuating lesion on portal venous phase images that was not associated with bile duct dilatation or delayed pooling of contrast material was rated 3. A hypoattenuating lesion on portal venous phase images was rated 4. A well-defined lesion measuring near water attenuation value (features typical of simple cysts) was rated 5 or 6.
MR criteria.—The following MR imaging criteria were used to evaluate HCC and CCA. A focal area of increased enhancement on arterial phase images was considered suggestive of HCC and rated 2 or 3. An additional criterion for diagnosing HCC was mild lesion hyperintensity on T2-weighted images. If a lesion showed both an arterial phase blush and mild hyperintensity on T2-weighted images, it was rated 1 or 2. Heterogeneous lesions or lesions that invaded the portal or hepatic vein were rated 1. For patients at increased risk of CCA, a lesion that showed arterial phase enhancement and mild hyperintensity on T2-weighted images was considered suggestive of CCA and rated 2 or 3. If the lesion also showed markedly delayed enhancement and/or a peribiliary distribution, a rating of 1 was assigned.
Lesions that were very bright on T2-weighted images with no internal enhancement (features typical of simple cysts) or with nodular peripheral enhancement on arterial and venous phase images with centripetal progression of enhancement on delayed postcontrast images (features typical of hemangiomas) were rated 6. Lesions that were nearly isointense to liver on T1- and T2-weighted images with marked uniform enhancement on arterial phase images (features typical of focal nodular hyperplasia) were rated 5 or 6. These lesions were rated 6 if they also had a central scar that was bright on T2-weighted images and showed delayed enhancement after contrast material administration. Tiny hepatic nodules (<1 cm) with signal characteristics indistinguishable from those of the remaining liver parenchyma and no evidence of arterial phase enhancement were considered regenerative nodules. Small nodules (<2 cm) that were hyperintense on T1-weighted images and hypointense on T2-weighted images were considered suggestive of dysplastic nodules. If these lesions showed no evidence of increased arterial phase enhancement, they were rated 4 or 5.
US criteria.—The following criteria were used to evaluate HCC or CCA. For HCC, a homogeneous or heterogeneous hypoechoic lesion, heterogeneous hyperechoic lesion, target lesion (a solid lesion with a hypoechoic halo), or lesion with internal color Doppler flow was rated 1–2. A homogeneous hyperechoic lesion or a lesion with the appearance of an atypical hemangioma (a solid lesion with a hyperechoic halo) was rated 3. A heterogeneous liver without a focal lesion was rated 4–5. An anechoic lesion with through transmission was rated 6. For CCA, focal bile duct wall thickening or a hypoechoic solid mass surrounding a bile duct was rated 1–2. Focal bile duct dilatation with abrupt termination and no mass was rated 3. Scattered isolated bile duct dilatation in the absence of a mass or duct wall thickening was rated 4–5. Absence of ductal abnormalities was rated 6.
PET criteria.—The following criteria were used to evaluate HCC and CCA. A discrete focus with increased tracer accumulation markedly greater than that in the hepatic parenchyma was rated 1–2. Focally increased FDG uptake, minimally greater than that in the liver was rated 3. Heterogeneous uptake in the hepatic parenchyma without a focal lesion was rated 4. No definite focally increased FDG uptake or FDG uptake lower than liver background was rated 5 or 6.
Pathologic Methods
Gross and histologic analyses of all explanted livers were performed by an experienced hepatobiliary pathologist (J.H.R.). The presence or absence of all lesions identified with one or more of the imaging tests (CT, MR imaging, US, or PET) was determined histologically on a lesion-by-lesion basis. All livers were initially sectioned at 5-mm or thinner intervals in a transverse plane at the site of a suspected lesion and routinely at 10-mm intervals in the coronal plane. If a lesion identified at an imaging test could not be demonstrated in the explant, representative histologic sections were obtained from the region of the liver that best corresponded to the lesion seen at the imaging test. In those patients in whom a lesion was not seen at an imaging test, the pathologic specimens were carefully reviewed for the presence of a hepatic malignancy.
Statistical Analysis
Patient-based analyses.—Agreement between the two observers for each imaging method was measured with the 
statistic. Differences between paired ordinal responses made by the observers for the same method were tested for statistical significance with marginal homogeneity tests (StatXact 4; Cytel Software, Cambridge, Mass). Receiver operating characteristic (ROC) curves were fitted to the rating scale data with an analysis program (ROCKIT 0.9B; C. Metz, University of Chicago, Ill), and the area under the ROC curve (Az) was calculated. However, it was not possible to test the Az scores for significant differences. This is because ROCKIT programs require either fully paired (correlated data) or unpaired (uncorrelated) rating scale data. To be fully paired, the same patients have to be imaged with all of the imaging tests, and the same observers have to read all of the images obtained from all of the tests. In our study, the same patients were imaged with all of the imaging tests, but two different observers interpreted the images produced by each test. We, therefore, could not use ROC analyses for fully paired data. Moreover, because we used images of the same patients for each test, our data were not unpaired, and we could not use ROC analyses for unpaired data. We could, however, test the rating scale data for differences from histologic truth. Because Shapiro-Wilk W tests indicated that the data were not normally distributed, we tested for differences from truth with nominal logistic regression analysis. We also used nominal logistic regression analysis to create probability plots for the rated responses. Effect likelihood ratio tests were used to determine the significance of the association between the rating responses and the histologic truth. Alpha was set at less than .05. For the rating scale data, we also calculated descriptive statistics, sensitivities, specificities, and 95% CIs. Shapiro-Wilk W tests, calculations, nominal logistic regression analyses, and effect likelihood ratio tests were performed with use of software (JMP; SAS Institute, Cary, NC).
Lesion-based analyses.—Some patients had multiple HCC nodules, and some had multiple false-positive lesions. To treat these multiple HCC nodules or multiple false-positive lesions as independent samples would violate the assumption of independence required for statistical testing. Data analysis on a lesion-by-lesion basis was, therefore, descriptive, with no statistical tests being performed. The data used for the lesion-by-lesion analysis were obtained from consensus readings between the two readers for each test (CT, MR imaging, and US) and were based on the six-point confidence scale. The six-point confidence scale was collapsed into a binary scale (1–3 = tumor; 4–6 = no tumor) for purposes of analysis.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
Patient-based Analyses
Data were available for all 25 patients with a patient-by-patient analysis. In nine patients, one or more lesions that were confirmed histologically to represent HCC either at the time of analysis of the explant (n = 5) or at sonographically guided biopsy (n = 4) were diagnosed. In regard to the blinded interpretations of the imaging studies, data were not available for the second MR observer for one of the positive cases and for the second US observer for two of the negative cases. The value was 0.67 ± 0.11 (standard error) for US observers, 0.35 ± 0.13 for CT observers, and 0.52 ± 0.12 for MR observers. None of the differences between observers were significant (P .32).
Table 1 contains the Az scores and 95% CIs for the observers. The marginal data for the PET observers were degenerate; thus, ROC curves could not be constructed for these observers. The Az scores for the other three tests ranged from a high of 0.93 for US (observer 1) to a low of 0.70 for CT (observer 2). ROC curves are plotted in Figure 1. Sensitivities were highest for US (0.89 for both observers) and lowest for PET, with sensitivities of 0 for both observers (Table 1). The next lowest sensitivities were for MR imaging (0.56 for observer 1 and 0.50 for observer 2). Specificities were highest for PET (0.88 for both observers). Specificities were similar for US, MR imaging, and CT, with means of 0.73, 0.72, and 0.72, respectively.
|
|
|
|
Lesion-based Analyses
Data were available for a lesion-by-lesion analysis in 22 of 25 patients (eight of whom had one or more HCC nodules). Among the three patients in whom data were not available, cassettes containing histologic material that would have allowed comparison with imaging findings were unavailable for review in one patient. In a second patient, no lesions were identified with any imaging study and no histologic diagnosis of HCC or CCA was made. The third patient (the ninth patient with HCC) was lost to follow-up for 15 months; all initial imaging findings were negative for HCC, but analysis of the explant indicated multiple HCC nodules. PET imaging results were not used in this analysis because of very poor performance of the test.
Overall, 68 lesions were identified with one or more imaging tests in these 22 patients; however, not all lesions were identified with every imaging test. Seven additional lesions shown at histologic analysis to represent HCC nodules were missed at all imaging studies. All seven of these nodules were 1 cm or smaller in size. Six of the seven HCC nodules were present in one patient; this patient was lost to follow-up for 15 months and almost certainly developed these tumor nodules in the interval between the imaging studies and the liver transplantation. The seven lesions that were missed with all tests are not discussed further.
A total of 18 HCC nodules were found at histologic analysis; CCA was not encountered in our study population. CT helped to correctly diagnose 13 of 18 tumor nodules; MR imaging, 14; and US, 13. Twelve of the 18 nodules were present in two patients. Of the five HCC nodules missed at CT, three nodules 1.5 cm or smaller (all in the same patient) were located in a region of extensive artifact caused by surgical clips. CT was correct in helping to diagnose three other tumor nodules in this same patient. Another 1.2-cm HCC nodule showed heterogeneous enhancement at the arterial phase and was interpreted as a regenerative nodule at CT. A fifth 2.8-cm tumor nodule was not apparent at CT, even in retrospect. Of the four HCC nodules missed at MR imaging, three (in the same patient) were 1.8 cm or smaller and were not visible, even in retrospect. Another was missed because images were not acquired through the inferior right lobe because of technologist error. All five HCC nodules missed at US were in the same patient and measured between 0.7 and 2.3 cm at gross pathologic examination. The echotexture of the liver was heterogeneous in this patient and most likely obscured the small nodules.
Lesion-by-lesion analysis revealed nine false-positive diagnoses of HCC nodules or CCA with CT, five with MR imaging, and nine with US. Of the nine lesions misdiagnosed at CT, four enhanced more than the liver at arterial and portal venous phases; three of these four lesions were subsequently shown at histologic analysis to represent peliosis, a biliary hamartoma, and a regenerative nodule. The fourth lesion had no histologic correlate. Three other lesions enhanced less than the liver at all phases. Histologic analysis revealed hemangiomas. An eighth lesion enhanced less than the liver on arterial and portal venous images and was shown to represent a regenerative nodule. The ninth lesion, observed in a patient with primary sclerosing cholangitis, measured 2 cm and was thought to represent a CCA at CT. It demonstrated less peripheral enhancement than did the liver at the arterial phase only (the lesion was not scanned at delayed imaging because of technologist error); at histologic analysis, an abscess was found.
Of the five false-positive diagnoses of HCC nodules at MR imaging, three (in the same patient) showed markedly increased signal intensity on T2-weighted images and enhanced on portal venous and delayed phase images. At histologic analysis, hemangiomas were found. Another lesion showed high signal intensity on T1-weighted images, was isointense on T2-weighted images, and was enhanced on hepatic arterial and portal venous phase images; however, there was no histologic correlate. The fifth lesion thought to represent HCC showed increased signal intensity on T1-weighted images and enhanced on hepatic arterial phase images only. Histologic analysis revealed a regenerative nodule.
Among the nine false-positive diagnoses of HCC nodules at US, six lesions in three patients were hypoechoic, with no internal flow at color Doppler imaging. Histologic analysis revealed regenerative nodules, three of which were hemorrhagic. Another lesion thought to represent HCC at US was hyperechoic and heterogeneous, but subsequent histologic analysis revealed a hemangioma. The last two lesions, which were adjacent to each other, were homogeneous and hyperechoic, but the larger of the two lesions showed internal color Doppler flow. Both were found to represent hemangiomas at histologic analysis.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
PET had little diagnostic value in this study population. There were only two cases that the two observers rated as being positive, and these cases were actually negative; thus, PET had a sensitivity of 0. With PET, as scores increased, there was a higher likelihood of the case being positive, which is just the opposite of what the rating scores were intended to mean. Agreement between observers was perfect for PET, but neither observer successfully identified a case as being positive.
Although our results showed that US outperformed CT and MR imaging on a patient-by-patient basis, these results should be interpreted cautiously given the small sample size, particularly for positive cases (n = 9). Nevertheless, our CT and MR imaging results are not entirely unexpected. In a recent study (3), CT was shown to be fairly insensitive in the detection of HCC and CCA. The authors reported low prospective (59%) and retrospective (68%) rates in the identification of HCC at triphasic helical CT in patients with cirrhosis, with even worse prospective (37%) and retrospective (44%) detection rates for individual tumor nodules (3). Their study was ideally designed in that it allowed assessment of the rate of tumor detection in a large surveillance population of patients with cirrhosis, but without known or suspected HCC, who were undergoing routine evaluation before liver transplantation. Surgical pathologic findings were tested for associations with CT in every case, regardless of the absence or presence of tumor, which allowed the inclusion of false-negative cases into the database and provided a more accurate assessment of the true sensitivity of CT (3). Our findings were similar; the sensitivities of CT for readers 1 and 2 were 0.67 and 0.56, respectively.
One important reason for the low sensitivity of CT in helping to detect HCCs relates to the lesion size (3). In one study, the mean size of the lesions detected at CT was 2.4 cm, compared with a mean size of 1.3 cm for those that were missed (3). When we examined our CT data on a lesion-by-lesion basis, we found that CT missed five of 18 HCC nodules. Three of the five nodules measured 1.5 cm or less and were in a region of extensive clip artifact. Another tumor nodule, which measured 2.8 cm, was not visible at CT, even in retrospect.
Artifacts, partial-volume averaging, respiratory motion, and the rare isoattenuating lesion at hepatic arterial and portal venous phases are additional factors that can lower the sensitivity of CT in the detection of HCCs. However, many false-negative CT findings occur because of the development and growth of tumors during the interval between the imaging study and transplantation (3). On the basis of the doubling time of HCC, it has been suggested that the CT surveillance interval for patients with cirrhosis who are awaiting liver transplantation should be 6 months (3). The interval between CT and liver transplantation was 2 months or less in the patients with the five lesions missed with CT in our study.
Authors of a more recent study (4) assessed whether double–arterial phase, multi–detector row helical CT imaging improved detection of HCC. The investigators found a 54% sensitivity for tumor detection in the early arterial phase, 78% in the late arterial phase, and 86% for the double–arterial phase imaging. The authors also noted an improvement in their positive predictive value with double–arterial phase imaging. However, 42 of the 51 patients who were entered into the study had HCC. Given the high prevalence of the disease in their study, the positive predictive value was almost certainly higher than would be expected in a surveillance population of liver transplant candidates.
Although authors of earlier studies reported very good sensitivities of MR imaging in the detection of HCC (5), our MR imaging results are more similar to those of another published study (6) that showed that state-of-the-art MR imaging is insensitive in the detection of small (<2 cm) HCC nodules. These authors compared their MR imaging findings with the pathologic findings in the liver explant on a lesion-by-lesion basis and found that MR imaging helped to detect only 11 of 19 HCC nodules, with a sensitivity of 53% (6). The mean size of their missed tumor nodules was 1.3 cm (6). In our study, MR imaging had a sensitivity of 0.56 and 0.50 for readers 1 and 2, respectively. Our lesion-by-lesion analysis demonstrated that MR imaging missed four of 18 HCC nodules; three of the four nodules were 1.8 cm or smaller. Lesion size is also an important factor in the depiction of tumor nodules with MR imaging.
Our high sensitivity in the detection of HCC with US on a patient-by-patient basis was unexpected (0.89 for both readers). However, on a lesion-by-lesion basis, US missed five of 18 HCC nodules, which is similar to the results of CT and MR imaging. All five tumor nodules missed at US were in the same patient and measured between 0.7 and 2.3 cm at pathologic evaluation. Had the missed nodules been in five different patients, the sensitivity of US would have been much lower on a patient-by-patient basis.
Findings from a prospective study (7) showed a sensitivity of only 43% (12 of 28) for the detection of HCC in patients with end-stage cirrhosis and 42% (29 of 69) for the detection of individual tumor nodules. During that study, all patients were examined by a sonographer, and approximately 65% were reexamined by a radiologist (in 10% of cases, the radiologist did not specialize in US), in contrast to our study in which a radiologist specializing in US examined every patient. Since that study, updated scanners, new advancements in transducer technology, improvements in image processing and resolution, and techniques such as harmonic imaging, real-time compound imaging, and intravenously administered contrast agents have increased the ability of US to depict subtle HCC nodules. However, harmonic and real-time compound imaging and intravenous contrast agents were not used during our study, and differences in technology and image resolution most likely account only in part for the high sensitivity of US shown in our study.
Authors of two retrospective studies (8,9) also reported a low sensitivity of US in the detection of HCCs. In one study (8), a sensitivity of only 38% (six of 16) for liver transplant candidates and 33% (six of 18) for the detection of individual HCC nodules was reported. However, patients were examined by radiologists who had variable expertise in US. In the other study (9), HCC was detected in 14 (67%) of 21 patients and in 21 (50%) of 40 individual tumor nodules. In that study, all patients were examined by a radiologist who specialized in hepatic imaging.
In contrast, authors of two older retrospective studies from Asia reported higher US detection rates of HCC (73%–84%) (10,11). In both studies, they compared US with CT and found that the detection rates were very similar for both tests. US is frequently used as a screening test in Asian countries, where the prevalence of hepatitis B and C is very high (10,11). In Japan, a screening program has been instituted for patients with chronic liver disease that includes measurement of -fetoprotein levels and performance of sonography (11). Perhaps gross pathologic differences in the appearance of HCC between Asian and non-Asian populations (higher frequency of a tumor capsule and fatty change in Asian populations) (24) explain, in part, the improved conspicuity of tumor nodules with US in Asian populations. Regardless, the detection of HCC nodules with US is dependent on and hampered by an inability to visualize the entire liver (especially the dome or a small left lobe), liver heterogeneity, isoechoic HCCs, small lesion size, large body habitus, inability to penetrate the entire liver because of fatty infiltration, operator expertise, and the infiltrative nature of some HCCs. A larger prospective study of a surveillance population of liver transplant candidates will need to be performed to better determine the true sensitivity of US in helping to detect HCCs.
Our lesion-based analyses demonstrated that false-positive diagnoses of HCC or CCA occurred with all imaging tests. There were nine false-positive diagnoses in five patients at CT; peliosis, biliary hamartoma, regenerative nodule, hemangioma, and abscess were found at histologic examination. There was no histologic correlate for one hyperattenuating lesion that may have been caused by a transient hepatic attenuation difference. Authors of several studies have emphasized the difficulty of distinguishing HCC from these lesions at CT (12,25,26). Authors of one study (25) reported that peliosis, hemangioma, focal fibrosis, and hepatic attenuation difference produced lesions that simulated the appearance of HCC in 11 of 429 patients with transplants. Hemangiomas have been reported to be rare in end-stage cirrhosis, owing to lesion obliteration by the fibrotic process (26,27). This process also alters the enhancement pattern (28) and most likely accounts for the misdiagnoses in our study.
There were five false-positive diagnoses of HCC in three patients at MR imaging. Three lesions diagnosed as HCC nodules were subsequently found to represent hemangiomas at histologic analysis. Like CT scans, hemangiomas on MR images may have an altered appearance due to lesion fibrosis and thus may become more difficult to recognize and differentiate from HCC (27). Another false-positive lesion represented a regenerative nodule at histologic analysis. Regenerative nodules at MR imaging are typically isointense on T1-weighted images, have a low signal intensity on T2-weighted images, and do not enhance (26,29). Only a small percentage of regenerative nodules are hyperintense on T1-weighted images (26). The MR imaging features of our false-positive lesion were far more characteristic of HCC than a regenerative nodule. There was no histologic correlate for another false-positive lesion; it may have been caused by a transient hepatic attenuation difference.
There were nine false-positive diagnoses of HCC in four patients at US. Six hypoechoic lesions were found to represent regenerative nodules at histologic analysis; three of those were hemorrhagic. Authors of very few studies have discussed the sonographic appearance of regenerative nodules. Findings from an earlier small study (30) in which sonographic findings were compared with gross pathologic findings in liver explants demonstrated that small regenerating nodules are isoechoic to surrounding parenchyma and not identifiable as discrete nodules. The authors concluded that only rarely do regenerating nodules simulate malignancy; two such cases that showed large regenerative nodules with focally altered echotexture suggestive of a malignancy have been reported (31). Findings from both of these studies were published in the early to mid 1980s. Although marked improvements in scanning equipment and advancements in imaging techniques may account in part for the increased visualization of regenerative nodules, it is most likely that hypoechoic regenerative nodules are not as uncommon as once thought.
Two other false-positive lesions seen at US were hyperechoic and showed internal color Doppler flow. These lesions represented hemangiomas at histologic examination. It is unusual for hemangiomas to show internal color Doppler flow, even with the use of intravenous microbubble contrast agents (32). Because HCC nodules are hypervascular and may appear hyperechoic (7,9,33), we rated these lesions as possibly HCC. Another false-positive lesion seen at US was heterogeneous but was subsequently shown to represent a hemangioma. HCC nodules when small are often hypoechoic, but as they grow, may become more heterogeneous (33). Although our lesion was small, we did not believe we could confidently exclude the diagnosis of HCC.
FDG PET has been very successful in the evaluation of metastatic disease of the liver (16). A recent meta-analysis demonstrated that FDG PET has high sensitivity (96%) and specificity (99%) in the detection of hepatic metastasis from colorectal cancer (34). However, FDG PET has been less successful in the detection of HCC. In one report, investigators examined 110 patients with a variety of liver tumors, including liver metastases from primary adenocarcinomas or sarcomas (n = 66), CCAs (n = 8), HCCs (n = 23), and benign hepatic lesions (n = 13) (16). PET correctly depicted all primary CCAs and hepatic metastatic lesions. However, only 16 of 23 HCCs demonstrated increased FDG uptake; the remaining seven had low FDG uptake as did the benign lesions (16). Similar results have been reported by other researchers (17,19,22). The degree of FDG uptake in HCCs has been shown to be related to the tumor grade; well-differentiated HCCs show low FDG uptake that is indistinguishable from that of benign lesions, and moderately or poorly differentiated HCCs demonstrate high FDG uptake that is similar to that of metastatic lesions (19,22).
In our study, PET demonstrated markedly heterogeneous and relatively poor FDG accumulation in the liver in all patients. This likely reflects poor hepatic function and distortion of the hepatic architecture in this study population and probably explains our inability to detect any HCC. Our results suggest that in patients with poor hepatic function, FDG PET is of little clinical value in the detection of HCC.
During the early and mid 1990s, authors of several studies compared various imaging tests in the diagnosis of HCCs (9–11,14); however, all studies were retrospective and used imaging techniques that are no longer used. Authors of a few recent studies have compared tests by using state-of-the-art imaging techniques (13,15). In one prospective study, 50 patients with 72 HCCs 3 cm or smaller in diameter were examined with MR imaging and CT (13). The authors suggested that arterial phase dynamic MR imaging is superior to helical CT in the detection of small HCCs; however, most lesions that were missed with either test were 2 cm or smaller. Authors of another study of 49 patients with 242 tumor nodules 3 cm or smaller suggested that both early-phase CT and dynamic MR imaging are the most sensitive means of detecting small HCCs; again, of the lesions missed by both tests, nearly all measured 2 cm or less (15). Our lesion-based analysis supports these findings; small lesions will be missed by all studies.
There are several limitations of our study. Our sample size was relatively small, and our sample had a higher prevalence of disease than would generally be found in patients being considered for liver transplantation. Another limitation is that CT, US, and MR imaging were performed between 1996 and 1998 with older equipment. This may have resulted in lower sensitivities for all imaging tests than would have occurred with more technologically advanced equipment that could depict smaller enhancing lesions or near isoechoic lesions. Since the time of our study, multi–detector row helical CT has emerged; this modality increased the spatial resolution of images and shortened the time for acquisition of data sets. Data sets can now be obtained specifically during the time frame of the hepatic arterial and portal venous phases rather than with some degree of overlap as occurs with single–detector row CT. Although this may improve the sensitivity in detecting small HCCs, it may also increase the number of false-positive findings and thus lower specificity.
Three-dimensional gradient-echo sequences have supplanted two-dimensional gradient-echo sequences for dynamic contrast-enhanced MR imaging of the liver in most centers. Three-dimensional sequences, which have effective section thicknesses of 2–3 mm, may improve the accuracy in the detection of small HCCs; however, thinner sections might also result in an increase in the number of false-positive findings, which would lower specificity. Finally, advancements in US, including real-time compound imaging and harmonic imaging, have markedly improved tissue contrast and spatial resolution and decreased artifacts. These technologic advancements would almost certainly improve the detection of subtle HCCs and CCAs.
A further limitation is that although each of the four imaging tests was interpreted blindly by two radiologists experienced with the test, for US the second reader reviewed a video tape of the US study, which probably biased the second reader toward seeing the lesions demonstrated by the first reader. However, the entire liver was scanned in longitudinal and transverse planes in a standardized fashion in an attempt to reduce this bias.
In conclusion, although on a patient-by-patient basis, US had the best diagnostic performance in helping to identify HCC, CT, MR imaging, and US performed similarly on a lesion-by-lesion basis. Small tumor nodules were the most common cause of missed HCCs with all tests. This is not unexpected and is well documented in the literature. In our study, small HCC nodules missed with one test were often seen with another. Therefore, there may be a diagnostic benefit from using more than one imaging test in the diagnosis of HCCs. False-positive findings occurred because of the presence of other pathologic lesions, such as regenerative nodules, hemangiomas, and peliosis that mimicked the imaging appearance of HCC. PET had the worst diagnostic performance, with no success in identifying any HCC, regardless of tumor size.
|
FOOTNOTES |
|---|
Abbreviations: Az = area under the ROC curve, CCA = cholangiocarcinoma, FDG = fluorodeoxyglucose, HCC = hepatocellular carcinoma, ROC = receiver operating characteristic
Author contributions: Guarantor of integrity of entire study, S.A.T.; study concepts and design, S.A.T., B.A.S., M.G.P., J.P.H., F.D., W.D.M., E.G.M., C.C.H., J.J.B., D.M.B.; literature research, S.A.T.; clinical studies, S.A.T.; data acquisition, S.A.T.; data analysis/interpretation, S.A.T., C.C.H.; statistical analysis, C.C.H.; manuscript preparation and definition of intellectual content, S.A.T.; manuscript revision/review, all authors; manuscript editing and final version approval, S.A.T., B.A.S.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  J. M. Willatt, H. K. Hussain, S. Adusumilli, and J. A. Marrero MR Imaging of Hepatocellular Carcinoma in the Cirrhotic Liver: Challenges and Controversies Radiology, May 1, 2008; 247(2): 311 - 330. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 V. W.-T. Lam, K. K. Ng, K. S.-H. Chok, T.-T. Cheung, J. Yuen, H. Tung, W.-K. Tso, S.-T. Fan, and R. T. P. Poon Incomplete Ablation After Radiofrequency Ablation of Hepatocellular Carcinoma: Analysis of Risk Factors and Prognostic Factors Ann. Surg. Oncol., March 1, 2008; 15(3): 782 - 790. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Ronzoni, D. Artioli, R. Scardina, L. Battistig, E. Minola, S. Sironi, and A. Vanzulli Role of MDCT in the Diagnosis of Hepatocellular Carcinoma in Patients with Cirrhosis Undergoing Orthotopic Liver Transplantation Am. J. Roentgenol., October 1, 2007; 189(4): 792 - 798. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Kuehl, P. Veit, S. J. Rosenbaum, A. Bockisch, and G. Antoch Can PET/CT Replace Separate Diagnostic CT for Cancer Imaging? Optimizing CT Protocols for Imaging Cancers of the Chest and Abdomen J. Nucl. Med., January 1, 2007; 48(1_suppl): 45S - 57S. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. A. Blake, A. Singh, B. N. Setty, J. Slattery, M. Kalra, M. M. Maher, D. V. Sahani, A. J. Fischman, and P. R. Mueller Pearls and Pitfalls in Interpretation of Abdominal and Pelvic PET-CT RadioGraphics, September 1, 2006; 26(5): 1335 - 1353. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C-S Li, R-C Chen, H-Y Tu, L-S Shih, T-A Zhang, J-M Lii, W-T Chen, S-J Duh, and L-C Chiang Imaging well-differentiated hepatocellular carcinoma with dynamic triple-phase helical computed tomography Br. J. Radiol., August 1, 2006; 79(944): 659 - 665. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Giorgio, G. Ferraioli, L. Tarantino, G. de Stefano, V. Scala, F. Scarano, C. Coppola, and L. Del Viscovo Contrast-Enhanced Sonographic Appearance of Hepatocellular Carcinoma in Patients with Cirrhosis: Comparison with Contrast-Enhanced Helical CT Appearance Am. J. Roentgenol., November 1, 2004; 183(5): 1319 - 1326. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. V. Sahani and S. P. Kalva Imaging the Liver Oncologist, July 1, 2004; 9(4): 385 - 397. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C.-H. Chang, G.-Y. Chau, W.-Y. Lui, S.-H. Tsay, K.-L. King, and C.-W. Wu Long-term Results of Hepatic Resection for Hepatocellular Carcinoma Originating From the Noncirrhotic Liver Arch Surg, March 1, 2004; 139(3): 320 - 325. [Abstract] [Full Text] [PDF]
|
|
|
|
