| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Liver Imaging Research Group and Department of Radiology, University of
California, San Diego, 408 Dickinson St., San Diego, CA 92103-8226.
2 Department of Radiology, Veterans Affairs Medical Center, San Diego, CA.
3 Present address: Department of Radiology, Tuen Mun Hospital, Hong Kong.
4 Present address: Department of Medical Imaging, Fundacion Santa Fe de Bogota,
University Hospital, Bogota, Colombia.
5 Biostatistics and Bioinformatics Division, Family and Preventive Medicine,
UCSD, San Diego, CA.
6 Neurosciences and Biostatistics and Bioinformatics, UCSD, San Diego, CA.
Received June 27, 2007;
accepted after revision September 4, 2007.
Address correspondence to C. Sirlin
(csirlin{at}ucsd.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. All multiphasic CT scans obtained at a cancer
center over a 2-year period were reviewed. In consensus, three authors
retrospectively identified 227 hepatic masses (
5 mm) in 55 noncirrhotic
patients that homogeneously hyperenhanced on arterial phase and then faded to
isoattenuation: 107 masses were malignant and 120 were benign; 37 patients had
benign and 18 patients had malignant masses. Two analytic approaches were
pursued: per lesion and per patient. For the per-lesion analysis, the mean
cross-sectional diameter of each mass was calculated and receiver operator
characteristics (ROC) were assessed. For the per-patient analysis, the maximum
lesion diameter was determined for each subject and logistic regression models
were used to predict lesion classification (benign vs malignant) based on
per-patient maximum lesion size and additional information.
RESULTS. Masses ranged from 5 to 84.5 mm. All 29 masses 22 mm
were benign. Size was a statistically significant classifier of benign versus
malignant lesions in the per-lesion analysis (p = 0.024, ROC area
under the curve) and a significant or trend-level predictor of tumor type in
the per-patient analysis (logistic regression p values of the
diameter coefficients: 0.01–0.07).
CONCLUSION. In noncirrhotic livers, relatively large size is a
significant or trend-level predictor for benign tumors. Homogeneously
hyperenhancing masses 22 mm that fade to isoattenuation are benign;
smaller masses may be malignant.
Keywords: contrast agent • contrast enhancement • CT • liver • mass
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The differential diagnosis for hyperenhancing masses can usually be narrowed by assessing contrast enhancement patterns. Ringlike arterial phase hyperenhancement is suggestive of liver metastasis from an extrahepatic primary neoplasm [1–4], although it may also be seen with liver abscess and occasionally around benign cysts if a vigorous bolus is given. Mosaic hyperenhancement also suggests malignancy, but this pattern is characteristic of hepatocellular carcinoma (HCC) [5–7]. Peripheral nodular enhancement with progressive coalescence of nodules indicates benign hemangioma. Homogeneous hyperenhancement with persistence on the delayed phase also suggests hemangioma [2, 8, 9]. In contrast, homogeneous hyperenhancement with rapid washout and delayed hypoattenuation relative to the liver raises concern for malignancy. This pattern is typical of HCC [10] but may also be observed in hypervascular liver metastases.
A diagnostically more challenging and nonspecific enhancement pattern is homogeneous hyperenhancement fading to isoattenuation. In this article, we refer to masses with this enhancement pattern as transient hepatic attenuating masses because of their superficial resemblance to transient hepatic attenuation differences [11–13], which also are characterized by homogeneous arterial phase hyperenhancement followed by rapid fading to isoattenuation. As opposed to transient hepatic attenuation differences, which are peripheral and triangular, transient hepatic attenuating masses are true masses and usually can be distinguished from transient hepatic attenuation differences by their mass effect and round or lobulated shape.
The differential diagnosis for transient hepatic attenuating masses includes focal nodular hyperplasia, hepatic adenoma, HCC, metastases, and other lesions. Thus, these masses may be benign or malignant [1, 6, 8, 9, 14, 15], and, based on our anecdotal experience, frequently cause diagnostic confusion. Therefore, identifying imaging and other features that help differentiate benign from malignant transient hepatic attenuating masses would help in the management of these problematic masses.
A commonly applied paradigm in oncology is that the larger the tumor, the more likely it is to be malignant. However, we hypothesized that in the normal liver, the opposite is true for transient hepatic attenuating masses—namely, the larger the transient hepatic attenuating mass, the more likely it is to be benign. The basis of this hypothesis is that growing malignant tumors tend to develop areas of internal ischemia or necrosis [16], features expected to cause textural heterogeneity on contrast-enhanced imaging. Thus, small malignant lesions may exhibit homogeneous hyperenhancement and manifest as transient hepatic attenuating masses, but large malignant lesions would be unlikely to have such an appearance. Benign tumors, on the other hand, may grow slowly without developing internal ischemia or necrosis and so may manifest as transient hepatic attenuating masses even if they are large.
The purpose of this study was to test the hypothesis that large size is a predictor for a benign transient hepatic attenuating mass in the noncirrhotic liver and to identify a lesion diameter threshold that excludes malignancy. We also sought to assess other predictors of benign transient hepatic attenuating masses (including number of lesions, patient age, and patient sex) and to generalize our results to populations with different prevalences of benign versus malignant conditions.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Selection of Transient Hepatic Attenuating Masses
One author (a research fellow with 2 years of experience) retrospectively
reviewed the diagnostic reports of all multiphasic 4-MDCT scans obtained for
clinical care over a 3-year period (2002–2004) to identify patients
without known cirrhosis or chronic hepatitis in whom at least one focal
hyperenhancing liver lesion was described on at least one CT examination.
In consensus, three radiologists (an attending radiologist with 9 years of
experience, an abdominal imaging clinical fellow with 5 years of abdominal CT
experience, and an abdominal imaging research fellow with 5 years of abdominal
CT experience) performed two image reviews, each review session spaced several
months apart. In the first session, the radiologists reviewed the multiphasic
CT examinations of the identified patients and selected a subset of patients
with lesions that met all of the following study criteria for a transient
hepatic attenuating mass: round, oval, or lobulated mass 5 mm in
diameter; homogeneous appearance on all imaging phases; hyperenhancement
relative to the liver on the hepatic artery phase; and fading of enhancement
to isoattenuation relative to the liver on portal venous and more delayed
phases.
Lesions were excluded if they were < 5 mm, had straight or geographic borders, were heterogeneous, had central scars, showed venous washout to become hypodense to the liver on portal venous and more delayed phases, or retained contrast enhancement in parallel with the blood pool on portal venous and more delayed images. During this first session, the radiologists were unaware of the final diagnosis, clinical history, or laboratory data, and they were not allowed to view any other imaging study.
In the second session, the radiologists re-reviewed each transient hepatic attenuating mass in conjunction with all available diagnostic data and, using the criteria described next, assigned to each selected mass a consensus diagnosis of malignant or benign.
Criteria for malignancy were histology results or, in patients with
histologically confirmed liver metastases or extrahepatic primary cancer,
liver lesion enlargement at follow-up imaging. Enlargement was defined as
5-mm diameter change for transient hepatic attenuating masses 
20 mm and a
10-mm change in diameter masses > 20 mm. Size changes were assessed on CT,
if available, and on MRI examinations otherwise.
Criteria for benignity were histology results, unchanged lesion size at follow-up imaging examinations performed a minimum of 24 months after CT, or benign lesion features on ancillary imaging examinations in patients with no history of malignancy in clinical records during a follow-up period of 48 months or longer after CT. The benign ancillary imaging feature used was inside-to-outside spoked-wheel enhancement pattern on perflexane lipid microspheres (Imagent, Alliance Pharmaceutical)–enhanced sonography [17]. Sonographic examinations were performed by a single sonographer in conjunction with a single abdominal radiologist (both with > 5 years of experience in the technique at the beginning of the study period).
All transient hepatic attenuating masses, up to a maximum of 10 per patient, were analyzed. In patients with more than 10 lesions, the 10 largest masses were selected (to reduce overrepresentation of patients with innumerable lesions). Most patients with more than 10 lesions had malignancy; thus, selecting the 10 largest lesions, rather than the smallest 10 or a random sample of 10, was a conservative decision against our hypothesis.
If patients had longitudinal multiphasic CT examinations showing growth of one or more transient hepatic attenuating masses, the examination showing the largest mass was selected. Because rapidly growing solid lesions are usually malignant, this was a conservative selection criterion against our hypothesis, designed to include in our analysis the largest possible malignant transient hepatic attenuating masses.
Multiphasic CT Technique
Multiphasic CT was performed on a 4-MDCT scanner (LightSpeed, GE
Healthcare) with 2.5-mm collimation and 5-mm reconstruction slice thickness.
According to our routine clinical protocol, images through the liver were
obtained before and after the bolus injection of 125 mL of iodinated contrast
medium (Optiray 320 [ioversol], Mallinckrodt) at 4–5 mL/s using a power
injector (CT 9000, Liebel-Flarsheim). Timing was achieved with bolus tracking
software (SmartPrep, GE Healthcare). Hepatic artery phase images were obtained
when aortic attenuation reached 170 H, and portal venous phase imaging began
30 seconds after arterial phase imaging was completed. In general, hepatic
artery phase imaging began at 30–35 seconds and portal venous phase at
60–70 seconds. More delayed images (3–5 minutes after injection)
were obtained at the discretion of the monitoring radiologist.
Image Review
Review sessions were performed using a 2,000 x 2,000 pixel resolution
gray-scale monitor on a PACS workstation (Impax, Agfa HealthCare).
Data Collection
The three radiologists electronically measured transient hepatic
attenuating masses to the nearest millimeter on the axial CT slice in which
the mass was largest. Orthogonal bidimensional measurements were made and
recorded on a spreadsheet. To assess longitudinal stability, masses were
remeasured on follow-up imaging examinations. Longitudinal changes in their
appearances were recorded.
The research fellow used the hospital computerized information system to collect and record demographics, clinical information, and pathology data. Prospectively dictated clinical radiology reports were reviewed and compared with the final consensus diagnosis.
Statistical Methods
Patient demographics and clinical data were summarized. Patients with
malignant and those with benign transient hepatic attenuating masses were
compared as to age (Wilcoxon's two-sample test) and sex (Fisher's exact test).
Each mean cross-sectional diameter of each mass was calculated. The number of
masses per patient was counted.
The distribution by size for each transient hepatic attenuating mass type (benign vs malignant) was compared descriptively. A size threshold that best discriminated benign from malignant was identified.
For statistical analyses, two approaches were pursued: per lesion and per patient. For the perlesion analysis, a receiver operator characteristic (ROC) curve was plotted for classification of benign and malignant transient hepatic attenuating masses on the basis of individual mass diameter. The area under the curve (AUC) was calculated and its significance assessed. For this assessment, bootstrap resampling (with 1,500 resampled data sets) was done on a per-patient basis to adjust for within-patient dependence of the data [18]. The AUC of the ROC curve for each resampled data set was calculated. A bias-corrected CI and a p value for the original AUC were then calculated from the bootstrap distribution [19]. The size cutoff that maximized accuracy was determined, and the accuracy at that cutoff was calculated.
|
|
|
For each regression model, the accuracy for classifying patients as having benign or malignant masses was determined. The p value of each predictor was computed.
Because our study population sample was obtained from a cancer center, benign lesions likely were underrepresented compared with the general population. According to sources in the literature and based on the latest Surveillance Epidemiology and End Results data released by the National Cancer Institute [20], among 100,000 people in the general population there should be approximately 900 cases of benign lesions that may manifest as a transient hepatic attenuating mass (focal nodular hyperplasia much more frequently than adenoma) [14, 15], three cases of HCC, and 60 metastases. Thus, although the actual prevalence of transient hepatic attenuating masses in the general population is unknown, it is highly likely that benign masses are considerably more common than malignant masses and that our models overestimated the probability of malignancy. Therefore, we performed a simulation analysis in which one of the per-patient models (logistic regression with per-patient maximum lesion diameter as the only predictor) was repeated with a series of hypothetic weights added to the fitting process. The weights varied the proportion of patients with benign masses from 20% (smaller than the observed proportion in our study sample) to 95% (greater than the observed proportion).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
One hundred twenty lesions were benign as determined by histology (n = 5), imaging stability for more than 24 months (n = 112), or compelling ancillary imaging findings in conjunction with clinical stability (n = 3). Two of the five histology-confirmed benign masses were biopsied because of growth during imaging follow-up (both were diagnosed as focal nodular hyperplasia); in contrast to growing malignant masses, which transformed into heterogeneous masses during their growth, the two growing benign masses did not change in character (Fig. 4A, 4B, 4C). The other three histology-confirmed benign masses were hepatic adenomas. Three masses were classified as benign focal nodular hyperplasia on the basis of an inside-to-outside spoked-wheel enhancement pattern on perflexane lipid microspheres–enhanced sonography (Fig. 5A, 5B, 5C) in conjunction with clinical stability.
|
|
|
|
|
|
Mass Size
Malignant masses ranged in size from 5 to 21.5 mm (mean, 9.9 mm; median, 9
mm); benign masses ranged from 5 to 84.5 mm (mean, 16.6 mm; median, 12 mm). As
shown in Figure 6A, all 29
masses with a diameter of 22 mm were benign. Of 198 masses with a
diameter 21.5 mm, 91 (46%) were benign and 107 (54%) were malignant.
|
|
Logistic Regression Models
In the three per-patient logistic regression models in which it was
entered, maximum per-patient transient hepatic attenuating mass diameter was a
significant or a trend-level predictor of tumor type: p values of the
diameter coefficient were 0.011 (imaging information model), 0.061 (univariate
model), and 0.073 (combined information model). Larger per-patient maximum
diameter was associated with benign lesions.
The number of transient hepatic attenuating masses was significant in both models in which it was entered as a predictor (p = 0.0007 and 0.001 for the diameter coefficient, fewer lesions associated with benign disease). Patient age was a significant predictor (p = 0.022 for the age coefficient) in the demographic model, with younger age associated with benign mass. Age was not a significant predictor (p = 0.32) in the combined information model. Patient sex was not a significant predictor of tumor type in either model in which it was entered (p = 0.20–0.26).
The univariate, imaging information, demographic, and combined information multivariate models had 66%, 80%, 66%, and 82% accuracy for classification, respectively.
The estimated probability curves for benign versus malignant masses based on perpatient maximum lesion diameter (univariate model) are shown in Figure 7. The estimated probability of benign masses increased as the underlying proportion of benign masses was varied from 20% to 95% (Fig. 7). For hypothetic proportions of 80% and 95% for benign masses, the probability of a benign mass exceeded 50% at all mass sizes.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
22 mm were
invariably benign; thus, a mass diameter of 22 mm had 100% positive
predictive value for benignity and excluded malignancy. By comparison, masses
21 mm were approximately equally likely to be benign or malignant. One plausible explanation for the association of larger mass size with benignity is that malignant lesions tend to become heterogeneous as they enlarge. This explanation is supported by our observation that all malignant masses followed up for more than 6 months not only grew but also changed in character from homogeneous to heterogeneous and thus converted from transient hepatic attenuating masses to nontransient hepatic attenuating masses. This suggests that malignant lesions may manifest as transient hepatic attenuating masses only if they are imaged during a relatively narrow time window in which they are large enough to be seen but not large enough to be ischemic or necrotic. By comparison, most benign transient hepatic attenuating masses in our study sample did not grow; the two that grew did not change in character.
Although size was a significant or trend-level predictor of malignancy or benignity, it did not by itself permit accurate classification for diagnostic purposes. Using size as the sole classifier, accuracy was 64% in the per-lesion analysis (at the size cutoff that maximized accuracy) and 66% in the per-patient analysis (univariate model). The poor diagnostic performance of size as a sole classifier is not unexpected: All lesions start small and then grow; thus, small masses may be benign or malignant, and size alone cannot differentiate small malignant from small benign masses.
Per-patient number of transient hepatic attenuating masses was a highly significant predictor of per-patient mass malignancy (fewer masses associated with benignity). Incorporating information on per-patient number of masses and mass size improved classification accuracy (accuracy of the imaging information model: 80%). Adding demographic information (age and sex) improved accuracy only modestly (accuracy of the combined information model: 82%). By comparison, a model that incorporated only demographic information had relatively low accuracy (66%).
Our population sample was derived from a cancer center with a high underlying prevalence of malignancy and probably overrepresented the frequency with which malignant transient hepatic attenuating masses would be seen in a general practice setting. In our simulation model, we increased the proportion of patients with benign transient hepatic attenuating masses from less than to more than that observed in our study. As expected, as the underlying proportion of patients with benign masses increased, the per-patient probability of benignity increased at each size threshold. In the two hypothetic populations in which the proportion of patients with benign masses was 80% or greater (probably more representative of the general population than our population sample), patients were more likely to have benign than malignant masses at all per-patient maximummass sizes. Thus, the hypothetic models suggest that in the general population all transient hepatic attenuating masses are likely to be benign, whether small or large. Nevertheless, conditional on a transient hepatic attenuating mass being present, the probability of benignity will still be higher for larger lesions.
Our study has several limitations. Its retrospective design may have introduced selection bias, but a prospective approach was impractical because of sample size and logistic considerations. Images were reconstructed at 5-mm intervals, and we included masses as small as 5 mm. It is possible that some 5to 10-mm masses would not have qualified as transient hepatic attenuating masses had thinner slices been obtained (e.g., minor heterogeneity may have been volume averaged), but 5-mm slice thickness is the routine radiology practice at our institution. Also, eliminating 5- to 10-mm lesions would not have meaningfully changed our results.
The selection criteria for transient hepatic attenuating masses were subjective and, because this was a preliminary test of hypothesis, reading was done by consensus. It would be meaningful to assess interobserver agreement in future studies. The readers were not blinded to the appearance of other liver lesions or to extrahepatic findings; thus, they often could determine whether the mass was malignant from ancillary imaging findings, which may have introduced interpretation bias. One solution for future studies would be to digitally mask all findings outside the lesion in question.
Reviewers did not record the presence, number, or characteristics of coincidental masses that did not qualify as transient hepatic attenuating masses; it is possible that adding information about coincidental masses would have improved the diagnostic performance of the logistic regression models.
Our study focused on assessment of transient hepatic attenuating masses in the noncirrhotic liver, in which primary liver cancers are relatively uncommon; and, in our population sample, all malignant masses were metastases to the liver. Thus, our study does not address assessment of masses in the cirrhotic liver. On the basis of our anecdotal experience, we believe that, in contrast to large transient hepatic attenuating masses in the normal liver, large masses in the cirrhotic liver should be regarded as suspicious for malignancy. This is in contrast to small masses in the cirrhotic liver, which are more likely to be benign abnormalities [21, 22] such as hypervascular regenerative nodules or pseudolesions caused by arteriovenous shunts.
By comparison, in the noncirrhotic liver, in which regenerative nodules are expected to be absent and in which arteriovenous shunts are uncommon, we found that 54% of small (< 22 mm) transient hepatic attenuating masses are malignant. The high probability of malignancy among small transient hepatic attenuating masses observed in our study, however, does not imply that small incidentally discovered masses should be regarded with suspicion in clinical practice. According to our simulation models, most transient hepatic attenuating masses seen in the general population are likely to be benign, regardless of their size. In addition, our logistic regression models indicate that adding information on per-patient number of masses improves classification accuracy. Thus, small masses need to be interpreted in the context of mass number; those that occur in the presence of multiple other masses are more likely to be malignant than those that occur in isolation. Although it was not addressed in our study, we believe that small masses need to be interpreted in the context of ancillary findings: Those that occur in the presence of suspicious nontransient hepatic attenuating mass lesions may be more likely to be malignant than those that occur in the absence of suspicious lesions. Further study is needed to confirm this conjecture.
By definition, transient hepatic attenuating masses appeared homogeneous on the arterial phase. Masses with minor degrees of arterial phase heterogeneity did not qualify and so were not analyzed. Anecdotally, we have observed that large masses with minor degrees of arterial phase heterogeneity also tend to be benign; thus, eventually it may be possible to relax the definition of transient hepatic attenuating mass to include masses with minor heterogeneity, but this will require further investigation.
Our study assessed only masses seen on CT. We have seen analogous masses—transient hepatic intensifying masses—on gadolinium-enhanced MRI (Fig. 4A, 4B, 4C). We hypothesize that size will be a significant predictor of lesion malignancy for transient hepatic intensifying masses and transient hepatic attenuation differences, although because of the high sensitivity of MRI to gadolinium enhancement, the lesion size diagnostic cutoff and the malignancy probability curves may differ from those reported here for CT.
Only a minority of lesions had histologic confirmation, and we relied on a noninvasive reference standard for lesions with no histology findings. Including lesions without histologic confirmation was necessary because focusing only on lesions with a definite reference standard may have biased our study toward unusual lesions that led to aggressive diagnostic follow-up.
Because histology was not available for all benign masses, our study does not assess whether size helps differentiate hepatic adenoma from focal nodular hyperplasia. This is important because hepatic adenoma may require resection due to risk of hemorrhage (especially if large), whereas focal nodular hyperplasia generally does not, regardless of size. Therefore, even though large transient hepatic attenuating masses are likely to be benign, other factors probably should be taken into account to differentiate benign entities, including medical history, risk factors, ancillary imaging findings, and additional diagnostic tests.
In clinical practice, large liver lesions other than simple cysts and hemangiomas tend to be aggressively investigated. Multiple diagnostic tests, including ancillary imaging examinations and biopsy, may be performed. Such tests increase health care costs, have risks, and, depending on how the differential diagnosis is formulated and how the situation is discussed with the patient, may cause profound patient anxiety. Contributing to patient anxiety and potential mismanagement is that large transient hepatic attenuating masses may be falsely interpreted at noninvasive imaging; in 24% of our patients with benign masses, the prospectively dictated clinical radiology report included only malignant conditions in the differential diagnosis.
The results from this study suggest that a conservative approach may be appropriate for large transient hepatic attenuating masses in the noncirrhotic liver because such lesions are likely to be benign. Additional testing may still be necessary to exclude malignancy and to differentiate benign lesions, but the differential diagnosis may be formulated, the situation may be discussed with the patient, and the workup may be directed in manners that minimize health care costs, risks associated with procedures, and patient anxiety.
We conclude that in this study population sample derived from a cancer center, in which malignant tumors likely are overrepresented, every transient hepatic attenuating mass with a diameter of 22 mm or greater was benign. Although additional studies are required to validate the 22-mm threshold, our findings suggest that large transient hepatic attenuating masses are likely to be benign, and the differential diagnosis should be formulated accordingly. Regression models can be used to estimate the probability of benignity at the patient level and can be adjusted to match the prevalence of benign versus malignant conditions in the target population.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
