| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Gastrointestinal Imaging |
1 From the Departments of Radiology (A.B., M.P.F., J.V.F., J.M.L., J.S.W., D.R.A., G.C., O.A., E.B.) and Biostatistics (W.L.), University of Pittsburgh Medical Center, 200 Lothrop St, Pittsburgh, PA 15213; and Department of Radiology, Spedali Civili Brescia, Italy (L.G.). Received April 17, 2001; revision requested June 5; revision received September 17; accepted October 10. Address correspondence to M.P.F. (e-mail: federle@pitt.edu).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
MATERIALS AND METHODS: Review of medical records at the University of Pittsburgh Medical Center identified patients with a total of 64 liver tumors that had central scars—including 29 cases of FNH, 20 fibrolamellar HCCs, and 15 large (>3.5 cm in diameter) hemangiomas—and with CT scans available for review. Retrospective review of these scans was performed individually by six radiologists who were blinded to the diagnosis, including two faculty abdominal radiologists, one abdominal imaging fellow, and three radiology residents. Individual performance was evaluated by means of receiver operating characteristic analysis, and interobserver agreement was measured by using the Cronbach 
. Individual CT findings that may allow differentiation of tumor types were identified with the Kruskal-Wallis test.
RESULTS: CT allowed good to excellent interobserver agreement in the diagnosis of tumor type and in recognition of differential findings among the three types. The individual accuracy of diagnosis was very good, with the average area under the receiver operating characteristic curve ranging from 0.81 to 0.90. Although the faculty radiologists performed the best, the differences in performance between the subgroups of readers and the levels of confidence in diagnosis were not statistically significant. The diagnosis of fibrolamellar HCC was the most accurate and had the highest sensitivity, followed by FNH and large hemangioma. Clinical and CT findings that were found to be statistically significant in differentiating tumor types were patient age and sex, tumor size larger than 10 cm, width of tumor scars, invasion of vessels, nodular centripetal enhancement, marked hyperattenuation on arterial phase images, lymphadenopathy, heterogeneity, extrahepatic metastases, surface lobulation, calcification, and isoattenuation with liver tissue on portal venous phase images.
CONCLUSION: CT allows accurate differentiation of the most common types of liver tumors with central scars, including FNH, fibrolamellar HCC, and large hemangioma.
© RSNA, 2002
Index terms: Diagnostic radiology, observer performance • Liver, focal nodular hyperplasia, 761.3198 • Liver, hemangioma, 761.3194 • Liver neoplasms, CT, 761.12111, 761.12114 • Liver neoplasms, diagnosis, 761.3194, 761.3198, 761.323
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
Regardless of their cause, scars within hepatic tumors are usually hypoattenuating compared with both the surrounding tumor and the normal liver parenchyma on unenhanced and contrast material–enhanced computed tomographic (CT) images. A characteristic feature of fibrous scar tissue within hepatic tumors is delayed and prolonged enhancement with intravenous contrast material (20–22). This feature differentiates central scarring from central necrosis, which is also seen in large tumors but does not demonstrate delayed contrast material retention.
Distinguishing FNH, fibrolamellar HCC, and hemangioma is important because benign lesions usually require no treatment, while fibrolamellar HCC requires aggressive surgical resection. Because all three types of tumors occur in noncirrhotic livers, have no associated serum markers, and are often classified as hypervascular, accurate diagnosis with the use of noninvasive tests has been considered difficult (5).
The purpose of our study was to determine the performance of radiologists with varying levels of expertise in the diagnosis of the most common types of liver tumors with central scars (ie, FNH, fibrolamellar HCC, and large hemangioma) by using specific CT findings that have been reported as typical of these tumors (1–19). Our hypothesis was that use of these CT criteria would allow accurate diagnosis of the type of tumor, thereby enabling appropriate management.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
The study was approved by our institutional review board, and informed patient consent was not required or obtained.
Review resulted in 90 patients with a diagnosis of FNH. Sixty-five of these patients had undergone abdominal CT. Of these 65 patients, 13 were excluded because of lack of presurgical imaging (n = 7) or because the images could not be located (n = 6). Ten patients without pathologic proof of FNH were excluded because they had insufficient imaging follow-up. Twenty-nine of the remaining 42 patients had FNH with a scar (Fig 1), and these patients were included in our study. Nine patients with FNH underwent resection, 12 patients underwent core biopsy, and eight patients had no pathologic proof of FNH, but their conditions were stable during a follow-up of 1–6 years (mean follow-up, 2.5 years). All of the excised areas of FNH had a distinct central scar and radiating fibrous septa.
|
|
Seventy-eight patients with pathologically proven fibrolamellar HCC were identified. Forty-two of these patients had undergone abdominal CT, but CT scans of 12 patients could not be located and were probably purged (images from 1994 and earlier have been purged at our institution). A central scar was demonstrated in 20 of the remaining 30 CT scans, and these were included in the study (Fig 2). Eleven of the 20 patients with fibrolamellar HCC underwent resection, and the remaining nine patients had core-biopsy proof of diagnosis. Histopathologic confirmation of fibrolamellar HCC was based on demonstration of large, polygonal, eosinophilic malignant hepatocytes that were surrounded by an abundant, collagenous fibrous stroma arranged in a lamellar distribution without underlying cirrhosis. There were 11 male and nine female patients (age range, 15–56 years; mean age, 29 years) with fibrolamellar HCC. Tumor size ranged from 50 to 190 mm (mean size, 140 mm; median size, 140 mm).
|
|
|
|
|
Sixty-four lesions were evaluated, including 29 cases of FNH, 20 fibrolamellar HCCs, and 15 hemangiomas. Fifty-two patients with 23 cases of FNH, 15 fibrolamellar HCCs, and 14 large hemangiomas underwent helical CT, and 12 patients with six cases of FNH, five fibrolamellar HCCs, and one hemangioma underwent conventional CT. Only one lesion per patient was analyzed; therefore, the number of tumors and patients was equivalent. Multiphase (ie, unenhanced, arterial phase, and portal venous phase) CT was performed in 22 patients with FNH, in five patients with fibrolamellar HCC, and in four patients with a large hemangioma. These five patients with fibrolamellar HCC and four of seven patients with FNH underwent only unenhanced and portal venous phase imaging. The three remaining patients with FNH and the four patients with large hemangioma underwent only portal venous phase imaging or late hepatic arterial phase imaging (some underwent delayed imaging). Delayed imaging of the liver had been performed in 12 patients with FNH, in seven patients with fibrolamellar HCC, and in 10 patients with large hemangioma.
A single tumor (ie, the only lesion per patient that contained a scar) was chosen for evaluation in each patient, although readers were not blinded to the presence of any additional lesions.
Image Evaluation
Each of the 64 CT scans were evaluated retrospectively and independently by six readers who had knowledge of the diagnosis of "liver tumor with central scar" but were blinded to the specific diagnosis and any clinical information other than patient age and sex. The six readers included two faculty radiologists (J.V.F. and J.M.L., each with 6 years of experience), one abdominal imaging fellow (J.S.W.), two senior residents (D.R.A. and G.C., each with 3 years of training), and one junior resident (O.A., with 2 years of training). The readers were not provided with any special training other than a short printed summary of what had been reported as "typical" findings of the three tumor types, including morphologic and radiologic tumor characteristics and other findings, such as lymphadenopathy and metastases (Table 1) (1–19).
|
The readers also evaluated large tumor size (>10 cm), tumor margins (sharp or ill defined), tumor surface (smooth or lobulated), and homogeneity. The tumor was considered homogeneous if it enhanced to the same degree throughout, with the exception of the scar and radiating septa. A capsule or pseudocapsule was considered present when a thin, curvilinear border surrounded the tumor and distinctly differed in attenuation from the adjacent liver parenchyma on unenhanced or enhanced images. The readers evaluated the presence of intralesional calcifications, as well as invasion of the hepatic vessels or bile ducts. The attenuation of the tumor relative to normal hepatic parenchyma was evaluated on unenhanced, hepatic arterial phase, portal venous phase, and delayed images when available. Isoattenuation with blood vessels and nodular centripetal enhancement were evaluated subjectively.
A tumor scar was defined as a central or eccentric area of attenuation that differed from the rest of the tumor on unenhanced or contrast-enhanced images. Scar characteristics that were recorded included scar size larger or smaller than 2 cm, delayed scar enhancement, and presence of radiating septa (defined as linear structures radiating peripherally from the scar).
Other findings, such as presence of extrahepatic metastases and lymphadenopathy (short-axis diameter of lymph nodes, >1.5 cm), were also evaluated.
Statistical Analysis
A biostatistician (W.L.) participated in the study design and review of the data. Interobserver agreement (ie, reliability) was evaluated with the Cronbach . We considered an value of more than 0.80 to represent excellent agreement (23) and values of 0.71–0.80 and 0.60–0.70 to represent good or acceptable agreement, respectively. Values of less than 0.60 were considered to represent poor agreement. Individual performance was evaluated according to the area under the receiver operating characteristic curve (Az), which was calculated with a nonparametric method (24,25). When making a comparison among readers, or when the results for readers were combined, we used a method that considered the correlation between readers, since all readers reviewed the same set of cases.
The percentage of correct diagnoses of tumor type (ie, sensitivity) was calculated for each reader. The tumor type rated with the highest degree of confidence was used as the diagnosis (eg, if a reader rated his diagnosis of hemangioma as "3," and fibrolamellar HCC received a "1" and FNH a "0," his diagnosis for that lesion was listed as "hemangioma"). The statistical significance of the differences in both sensitivity and degree of confidence for diagnosis between pairs of readers was evaluated by using the McNemar test (26) (Table 2). The McNemar test compares the dichotomous outcome of whether reader 1 correctly diagnoses each case with the dichotomous outcome of whether reader 2 correctly diagnoses each case.
|
The importance of other factors (ie, patient age and sex) in differentiating various tumor types was also assessed by using one-way ANOVA for patient age and the 
2 test (applied to the appropriate contingency table) for patient sex. We did not evaluate clinical factors such as liver function tests, tumor markers, clinical symptoms, or signs that may also be important in differentiating these tumors.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
|
The individual performance of readers in the diagnosis of various tumors was evaluated according to Az values, which were calculated by using nonparametric analysis (Table 4) and were found to be accurate for all readers (average Az, 0.81–0.90). The faculty readers were slightly more accurate than the other readers, with an average Az of 0.87 and 0.90. The Az of the residents and fellow were 0.81–0.88 and 0.87, respectively. The difference in accuracy of diagnosis between subgroups of readers was not statistically significant. Fibrolamellar HCC was diagnosed most accurately by all readers, followed by large hemangioma and FNH. We also generated an average Az value to evaluate the performance of the group as a whole, and found it to be very accurate, with an average Az of 0.91. Fibrolamellar HCC was diagnosed most accurately by the group (Az = 0.91), followed by hemangioma (Az = 0.86) and FNH (Az = 0.83).
|
2 test and one-way ANOVA with P values of less than .001.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
We were encouraged to find that there was excellent interobserver agreement in diagnosing the tumors and in recognizing almost all of the CT findings that were evaluated. All of the radiologists performed well, and there was no significant difference in performance between individuals or groups (ie, resident, fellow, and faculty groups).
Findings that were found to be statistically significant in differentiating the three tumor types were related to epidemiologic and imaging factors, including patient age and sex, tumor and scar characteristics, tumor enhancement patterns, and other findings.
More than 80%–95% of patients with FNH are women (3,5). A female predominance for hemangioma has been cited as well, although this was not evident in our population of patients who had large hemangiomas with central scarring (14). Fibrolamellar HCC shows no particular sex predominance, but it is characteristically discovered in adolescents and young adults, as shown in our study (9–11). FNH can be discovered at any age, although it has been reported most commonly in young women. The mean age of FNH patients in our study was 39 years, which closely matches that reported by other investigators (1–5). Large hemangiomas are usually detected in older patients, with a mean age of 54 years in our study and 45 years in another report (16).
Tumor Morphology
We found a few important distinguishing tumoral imaging features, including large tumor size (>10 cm), tumor surface lobulation, and intratumoral calcifications. FNH usually appears as a lesion smaller than 5 cm, whereas fibrolamellar HCCs and hemangiomas with central scars tend to be larger (1,2,4,5). Fibrolamellar HCC varies in size from 5 to 20 cm, with an average reported size of 13–14 cm (9–12) (average size was 14 cm in our study). Large or giant hepatic hemangiomas are defined as larger than 4 cm (17), 6 cm (18), or even 10 cm by various authors (19), with an average size of 11 cm (as in our study). The tumor surface can be an important distinguishing feature among the three tumor types. Fibrolamellar HCC usually has a lobulated margin (9,10,12) (in up to 65% of cases, according to Ichikawa et al [9]), whereas FNH commonly has a well-marginated, smooth surface (1,2,4), especially in tumors smaller than 3 cm (1).
Hepatic hemangiomas tend to have a rounded lobular margin (15). Calcification is typical of fibrolamellar HCC, with a reported incidence of 35%–68% (9–12). Calcifications may be punctate, nodular, or stellate and are usually small (<5 mm), few (one to three in number), and almost always located near the center of the tumor (9,10). Calcification is rarely seen in FNH. In a prior report (1) in which 124 FNH tumors were evaluated with CT, only one lesion had a calcification (0.8%). In another study (7) in which 357 cases of FNH were evaluated with CT, MR imaging, and ultrasonography, five lesions (1.4%) had a calcification. Calcifications have rarely been reported with sclerosed or hyalinized and giant hemangiomas (14,27,28). The calcifications may be marginal or central; large and coarse; or multiple, small, and punctate (eg, phleboliths). Although intralesional calcifications may be seen with any of the three types of tumors, their rarity in FNH and hemangioma and high frequency in fibrolamellar HCC make this an important differentiating feature, especially when evaluated with regard to other distinguishing imaging features.
Other tumor characteristics, including margins, presence of radiating fibrous septa or pseudocapsules, and number of lesions, did not prove as useful in distinguishing the three types of tumors.
Scar Characteristics
Fibrolamellar HCC demonstrates a central scar in up to 20%–71% of cases (9–12). It is typically large and may be broad or stellate, eccentric or central. Although some authors reported that the scar does not enhance on delayed images and that absence of scar enhancement can be used to distinguish fibrolamellar HCC from FNH (13), results of larger and more recent studies indicate that the scar retains contrast material and becomes hyperattenuating compared with the rest of the tumor in 25%–56% of cases (9,10). A central scar is histologically present in almost all patients with FNH. However, it may be subtle and extremely small and is identified with the use of CT in 30%–50% of cases (1,3,4,6). Presence of a central scar is clearly related to lesion size. Thirty-five percent of small areas of FNH (<3 cm) and 65% of large areas of FNH (>3 cm) demonstrated the scar in a recent article (1). The scar is typically thin and small and is hypoattenuating compared with the rest of the FNH on unenhanced, hepatic arterial phase, and portal venous phase images. On delayed images, the scar often becomes hyperattenuating because of contrast material retention (1,2,5,6). Large hemangiomas usually demonstrate cleftlike central or eccentric areas of scarring that constitute areas of fibrosis, sclerosis (ie, hyalinization), cystic degeneration, or thrombosis. These areas have a variable shape and size and usually do not demonstrate delayed contrast enhancement (16).
The only scar characteristic that we found to be a statistically significant differentiating feature was a scar size of larger than 2 cm, which indicated fibrolamellar HCC and sometimes hemangioma. Delayed scar enhancement and presence of fibrous radiating septa from the scar were not statistically significant features.
Tumor Enhancement Patterns
Tumor heterogeneity was found to be an important differentiating feature. Tumor heterogeneity is typical of fibrolamellar HCC and stems from its mixed content of eosinophilic neoplastic cells interspersed in collagen and fibrotic tissues and from intratumoral necrosis and hemorrhage (9,10,12,13). Fibrolamellar HCC is almost always heterogeneous (in 90% of cases, according to Ichikawa et al [9]) on unenhanced images and demonstrates heterogeneous enhancement on hepatic arterial phase images (9,10,12,13) (in 80% of cases, according to Ichikawa et al [9]). FNH, on the contrary, is usually homogeneous (1–5) and isoattenuating (40%–48%) to slightly hypoattenuating (42%–57%) (1,4) compared with normal liver tissue on unenhanced images. FNH demonstrates a homogeneous, immediate, bright enhancement in nearly all cases (89%–100%), with the exception of the central scar (1,2,4). Large hemangiomas are also relatively heterogeneous on unenhanced and contrast-enhanced images, but they demonstrate a characteristic enhancement pattern. This pattern of centripetal nodular enhancement with progressive filling (29,30), as well as isoattenuation with the blood vessels on unenhanced and contrast-enhanced images (30), is particular to hemangioma.
The last enhancement feature that we found to be a significant differentiating feature was isoattenuation with normal liver tissue on portal venous phase images. The enhancement of fibrolamellar HCC becomes less heterogeneous and pronounced on portal venous phase (as opposed to arterial phase) images, and the tumor is most commonly isoattenuating or hypoattenuating compared with normal liver tissue. Because of its prominent fibrous component, it may occasionally be hyperattenuating compared with normal liver tissue (9,10). FNH demonstrates homogeneous but decreased enhancement on portal venous phase (as opposed to arterial phase) images and is isoattenuating with normal liver tissue in most cases (72%–89%) (1,4,5), although it may be hypoattenuating or hyperattenuating (1,3,5). Hepatic hemangiomas are easily diagnosed on portal venous phase images according to their characteristic enhancement pattern. They usually demonstrate progressive filling and become hyperattenuating compared with liver tissue (paralleling hepatic vessels) in most cases, with the exception of the central area of scarring.
Other Findings
Patients with fibrolamellar HCC frequently have advanced disease at the time of initial presentation. Lymphadenopathy has been reported in 50%–70% of cases (9,10) and is not a feature of hemangiomas or FNH—therefore, it is an important distinguishing feature. Lymphadenopathy almost always involves the hepatic hilum and hepatoduodenal ligament and may be extensive in 60% of cases, involving multiple abdominal and retroperitoneal sites (9,10). Extrahepatic metastases to the lungs, peritoneum, or adjacent organs occur less commonly (20%) (10). Invasion of the hepatic vessels or bile ducts was also found to be an important differentiating feature, but it may be seen in less than 5% of cases of fibrolamellar HCC (9). Venous thrombosis may result from compression of the portal or hepatic veins by an adjacent tumor, which simulates vascular invasion. Portal vein thrombosis may be seen with fibrolamellar HCC (7) but has also been reported rarely with large hemangiomas (14).
Our study is limited to a degree by its retrospective nature, which we believed was unavoidable due to the relative rarity of the tumors. We believe that we were able to minimize most biases by including relatively large numbers of tumors and by avoiding exclusion of cases, other than the few that were not available for review due to purging. We believe that the inclusion of residents and a fellow as readers makes our findings and conclusions more applicable to individuals of various experiential levels.
In conclusion, we found that CT allows accurate differentiation of FNH, fibrolamellar HCC, and large hepatic hemangioma by radiologists with varying levels of experience and expertise.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Author contributions: Guarantor of integrity of entire study, M.P.F.; study concepts, A.B., M.P.F.; study design, M.P.F.; literature research, A.B.; clinical studies, all authors; data acquisition, A.B.; data analysis/interpretation, A.B., W.L.; statistical analysis, W.L.; manuscript preparation, A.B., M.P.F.; manuscript definition of intellectual content and editing, M.P.F.; manuscript revision/review and final version approval, all authors.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODS RESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  M. E. Lockhart, F. N. Tessler, C. L. Canon, J. K. Smith, M. C. Larrison, N. S. Fineberg, B. P. Roy, and R. H. Clements Internal Hernia After Gastric Bypass: Sensitivity and Specificity of Seven CT Signs with Surgical Correlation and Controls Am. J. Roentgenol., March 1, 2007; 188(3): 745 - 750. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Z.-C. Zeng, Z.-Y. Tang, J. Fan, J. Zhou, L.-X. Qin, S.-L. Ye, H.-C. Sun, B.-L. Wang, J.-Y. Zhang, Y. Yu, et al. Radiation Therapy for Adrenal Gland Metastases from Hepatocellular Carcinoma Jpn. J. Clin. Oncol., February 1, 2005; 35(2): 61 - 67. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| RADIOLOGY | RADIOGRAPHICS | RSNA JOURNALS ONLINE |
