| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Surgery, University of California at San Francisco, San
Francisco, CA.
2 Department of Radiology, Division of Abdominal Imaging, University of
California at San Francisco, 505 Parnassus Ave., Rm. M-372, Box 0628, San
Francisco, CA 94143-0628.
3 Department of Epidemiology and Biostatistics, University of California at San
Francisco, San Francisco, CA.
Received April 3, 2007;
accepted after revision May 19, 2007.
Address correspondence to F. V. Coakley
(fergus.coakley{at}radiology.ucsf.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. We retrospectively identified 58 patients with histopathologically proven cystic pancreatic masses at our institution who underwent preoperative CT (n =40), MRI (n = 6), or both (n = 12). Two radiologists independently recorded their leading diagnoses with levels of diagnostic certainty (0–100%), their estimates of overall likelihood of malignancy (0–100%), and the morphologic characteristics of the tumors. Data were analyzed to determine relative accuracy in the diagnosis of malignancy, relationship between diagnostic certainty and accuracy, and frequency of malignancy in unilocular thin-walled cysts smaller than 4 cm.
RESULTS. Twenty-one (36%) of 58 masses were malignant. CT and MRI were equally accurate in establishing the diagnosis of malignancy (area under the receiver operating characteristic curve [Az] = 0.91 and 0.85 for reviewers 1 and 2 at MRI vs 0.82 and 0.76 at CT, respectively; p > 0.05). The leading diagnosis given by reviewers 1 and 2 was correct in 46% (32/70) and 43% (30/70) of the studies, respectively. When reviewer diagnostic certainty was 90% or more, the corresponding values were not significantly (p > 0.05) improved at 55% (12/22) and 48% (10/21), respectively. Two (15%) of 13 unilocular thin-walled cysts smaller than 4 cm were frankly malignant.
CONCLUSION. CT and MRI are reasonably and similarly accurate in the characterization of cystic pancreatic masses as benign or malignant; limitations include a substantial rate of misdiagnosis even when reviewer certainty is high and a moderate frequency of malignancy in small morphologically benign-appearing cysts.
Keywords: CT • cystic pancreatic mass • histology–imaging correlation • MRI • pancreas • pancreatic cancer
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Noninvasive characterization of cystic pancreatic neoplasms continues to rely principally on CT and MRI, although published results have been mixed [9, 10]. Many case series describe the "classic" radiologic characteristics of various cystic pancreatic neoplasms [5, 11, 12], but few have addressed, to our knowledge, the diagnostic accuracy of cross-sectional imaging for cystic pancreatic lesions taken as a whole, investigated limiting factors, or compared the relative accuracy of CT and MRI. Proponents of MRI argue that its higher soft-tissue contrast and multiplanar capability result in superior differentiation among cystic neoplasms [13], but others have found that MRI adds little to CT [14, 15]. Whereas newer techniques, such as PET and endoscopic sonography with cyst aspiration [16], have been used to distinguish benign from malignant lesions at major referral centers, these results have not been widely duplicated [17], and the reality is that CT and MRI remain the primary means of assessment of cystic pancreatic masses in the community. Therefore, we undertook this study to determine the role and relative accuracy of CT and MRI in the characterization of cystic pancreatic masses.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The 58 patients in the study group consisted of 20 males and 38 females, two of whom were children (a 6-year-old boy with a gastric duplication cyst and a 14-year-old girl with a solid and cystic papillary epithelial neoplasm), with a mean age of 66 years (age range, 6–89 years). Of the 58 patients, 40 underwent CT, six underwent MRI, and 12 underwent both. Ten of 58 lesions were discovered incidentally, whereas the remaining 48 cases were discovered during an investigation of abdominal symptoms such as pain, discomfort, vomiting, or weight loss. The final histopathologic diagnoses are listed in Table 1 and were established by surgical resection (n = 45) or core needle biopsy (n = 13).
|
CT Technique
CT studies were performed using helical single-detector CT (n =
19; HiSpeed Advantage, GE Healthcare) or 4- or 16-MDCT (n = 33;
LightSpeed, GE Healthcare). All patients received 150 mL of 60% iodinated
contrast material (iohexol, Omnipaque 350, Nycomed Amersham) administered IV
with a power injector at a rate of 4 mL/s and 800 mL of diatrizoate meglumine
(Hypaque, Nycomed Amersham) administered orally. All scans included
acquisition of portal venous phase images with 5- to 7-mm collimation at 70
seconds after initiation of IV contrast material injection. In 41 patients,
arterial phase images were also acquired with 2.5- to 5-mm collimation at 30
seconds after initiation of IV contrast material injection.
MR Technique
MRI was performed on 1.5-T MR scanners (Signa, GE Healthcare) using a torso
phase array coil. The MR protocol included the following breath-hold
sequences: axial in- and out-of-phase gradient-recalled echo T1-weighted (TR
range/in-phase TE, out-of-phase TE, 90–150/4.2, 2.1; flip angle,
75°; slice thickness, 8 mm; interslice gap, 1 mm), axial fast spin-echo
(FSE) T2-weighted with fat saturation (TR/TE, 4,000/100; echo-train length, 8;
slice thickness, 6 mm; interslice gap, 1 mm), axial and coronal FSE
single-shot T2-weighted (infinite/100; slice thickness, 4 mm; interslice gap,
0 mm), coronal FSE single-shot heavily T2-weighted MR cholangiography with fat
saturation (infinite/180; slice thickness, 4 mm; interslice gap, 0 mm), and
axial 2D spoiled gradient-echo T1-weighted with fat saturation (150/4.2; flip
angle, 75°; slice thickness, 6–8 mm; interslice gap, 0–1 mm)
after dynamic IV gadodiamide injection (Omniscan, GE Healthcare) (0.1 mmol/kg
of body weight through a power injector). Initial contrast-enhanced images
were obtained 20–30 seconds after injection of the gadodiamide bolus,
and repeat images were obtained at 1, 2, and 3 minutes after injection.
Image Interpretation
Two attending radiologists with subspecialist interest in abdominal imaging
independently reviewed all CT and MR studies during separate sessions on a
PACS workstation (Impax, Agfa). The reviewers had 8 and 4 years' experience,
respectively, in the subspecialty of abdominal imaging at the time of the
study. To prevent learning or recall bias, reviews of CT and MR studies from
the same patient were separated by an interval of at least 1 month. Reviewers
were aware that patients had a cystic pancreatic mass, but they were unaware
of all other histopathologic and clinical data other than demographic data of
age and sex (routinely displayed on our PACS).
Each reviewer recorded specific morphologic tumor characteristics (Appendix 1), their leading diagnosis or diagnoses, and the overall likelihood of malignancy. Reviewers were allowed to list up to four specific leading diagnoses and were asked to assign each listed diagnosis a percentage probability, and the assigned probabilities had to total 100%. Reviewers were given a choice of specific histologic diagnostic options (pseudocyst, mucinous cystic neoplasm, serous cystadenoma, intraductal papillary mucinous neoplasm, duct ectasia associated with chronic pancreatitis, solid and cystic pancreatic tumor), but they could also choose from less specific diagnoses (benign simple cyst or cystic malignancy) or enter any alternative diagnosis that they favored. In addition, reviewers were asked to assign a level of certainty to their leading diagnosis from 0% to 100%. So, for example, if a reviewer was absolutely certain that a cystic mass was a pseudocyst, then that diagnosis would be given a probability of 100% and a level of certainty of 100%.
|
Reviewers used their expert judgment and knowledge of the published imaging criteria for the characterization of cystic pancreatic masses [5, 9–16]. As such, the criteria used in any individual patient by either reviewer were not recorded. However, in general, a diagnosis of pseudocyst was considered for unilocular lesions, particularly if exophytic and accompanied by signs of pancreatitis. A diagnosis of mucinous cystic neoplasm was considered for multicystic lesions with a lobulated appearance due to the presence of several locules of large size. Serous cystadenoma was considered for multicystic lesions composed of innumerable small locules, particularly if accompanied by a central scar or calcification. Intraductal papillary mucinous neoplasm was considered when a cystic lesion appeared to be composed of dilated or intercommunicating tubes, particularly if found in communication with the pancreatic duct or accompanied by main duct dilatation and a patulous ampulla of Vater. In addition to these morphologic factors, reviewers were aware of patient's age and sex, and these factors may also have influenced the choice and weighting of a given diagnosis.
|
|
|
The accuracy of CT and MRI in the diagnosis of malignancy was compared by
calculating the area under the receiver operating characteristic curve
(Az) for each technique; for the purposes of this
analysis, premalignant lesions were categorized as benign. The relationship
between diagnostic certainty and accuracy was determined using exact
2 and binomial methods. In particular, we examined whether the
leading diagnosis was more likely to be correct when the level of reviewer
certainty in the leading diagnosis was 90% or greater. Based on the approach
to characterizing cystic masses in other abdominal organs such as the kidney
and ovary, we specifically recorded the frequency of malignancy in
morphologically benign-appearing lesions, which we defined as small (4 cm or
less in diameter), unilocular, and thin-walled (3 mm or less) cysts.
Multivariate analysis (manual stepwise regression using backward elimination) was used to investigate morphologic characteristics as predictors of malignancy and premalignancy. The relationship between key morphologic characteristics and specific diagnoses was analyzed with the analysis of variance for continuous variables and the chisquare test for discrete characteristics. Interobserver variability was analyzed with Pearson's correlation coefficient (for continuous variables) and McNemar chi-square test (for binary variables). All statistical analyses were performed using statistics software packages (Stata version 8.0, Stata Corp.; and StatXact, Cytel Software).
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
With respect to specific histopathologic characterization, the leading diagnosis given by reviewer 1 and by reviewer 2 was correct in 46% (32/70) and 43% (30/70) of the imaging studies, respectively. Of note, both reviewers offered the same leading diagnosis in 28 (40%) of 70 cases, and in those cases the diagnosis was correct in 16 (57%). Reviewer 1 included the correct diagnosis in 61% (43/70) of imaging studies when his two leading diagnoses were considered and in 66% (46/70) of scans when his three leading diagnoses were considered. For reviewer 2, the corresponding results were 64% (45/70) and 67% (47/70).
With respect to specific histopathologic characterization when reviewer diagnostic certainty was 90% or more, the leading diagnosis given by reviewer 1 and 2 was correct in 55% (12/22) and 48% (10/21) of cases, which was not significantly (p > 0.05) improved from the respective accuracies of 46% (32/70) and 43% (30/70) for the characterization at any level of certainty (Fig. 2).
Two (15%) of 13 unilocular thin-walled cysts smaller than 4 cm in diameter were malignant (Fig. 3) and consisted of one mucin-producing ductal adenocarcinoma and one mucinous cystadenocarcinoma. Three (23%) of the 13 unilocular thin-walled cysts less than 4 cm in diameter were premalignant mucinous cystadenomas.
|
The final histopathologic diagnosis and leading diagnoses with level of probability offered by both reviewers in the 70 CT and MRI studies entered into this study are shown in Table 2. The data shown in Table 2 suggest that none of the individual final histopathologic diagnoses was reached with superior accuracy and that unexpected diagnoses occurred even in cases with the same leading diagnosis offered by both reviewers (Figs. 4A, 4B, 5, and 6), although the small number of cases per diagnosis limited any meaningful statistical confirmation.
|
|
|
|
|
The overall diagnostic accuracy and likelihood of predicting malignancy were not significantly different between the two reviewers (McNemar chi-square test: p = 0.86 and 0.63, respectively). Interobserver agreement for continuous and binary variables was good, with Pearson's correlation coefficient for continuous morphologic variables ranging from 0.73 to 0.90. Multivariate analysis of individual morphologic characteristics as predictors of malignancy and of specific diagnoses revealed no useful predictors.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Second, our study shows only limited accuracy for the specific diagnosis of cystic pancreatic mass based on CT or MRI appearance even when reviewers are convinced of the diagnosis. This finding argues against the view that there is a subset of cystic pancreatic masses with a "typical" appearance, at least in a surgical series. Our finding that the reviewers were correct in only 57% of the cases for which they both offered the same leading diagnosis also supports this interpretation. This result suggests that radiologists should approach the specific characterization of cystic pancreatic masses with a substantial degree of humility. The differential diagnosis of "likely pseudocyst or cystic neoplasm" may be maddeningly vague and broad to a referring physician, but may often be the most honest and appropriate radiologic impression to offer. The results of our multivariate analysis for morphologic predictors of specific diagnoses also support such a guarded approach. Despite having analyzed an array of imaging characteristics, we found no useful predictors, presumably due the heterogeneity among cystic lesions and overlap in imaging characteristics. The only specific condition that appeared to be diagnosed with a high positive predictive value by both reviewers was intraductal papillary mucinous neoplasm, although small numbers precluded meaningful statistical confirmation of this observation.
Third, one of our primary objectives in conducting this study was to compare the accuracy of CT and MRI in the characterization of cystic pancreatic masses as benign or malignant. Proponents of MRI argue that its higher soft-tissue contrast and multiplanar capability result in better differentiation among cystic neoplasms [13, 15] and that MR cholangiopancreatography can better depict the fluid content of a cystic mass and communication with the pancreatic duct [20–23]. Others counter that MRI offers poorer spatial resolution and that reformation of helical CT data can reveal similar detail with respect to ductal communication [24]. Our results suggest MRI offers little to no incremental benefit. It is true that only a limited number of patients underwent MRI and that there was a trend for both reviewers to be more accurate at MRI, so a larger study might have shown a significant benefit for MRI. However, it is difficult to imagine that our study would have missed a difference between the techniques of sufficient magnitude to be of substantive clinical importance.
Our study results are largely concordant with those of prior reports. Blinded retrospective analyses of CT for the characterization of cystic pancreatic masses have met with mixed success to our knowledge and, in most studies, have been limited by inclusion of only serous and mucinous tumors [9, 10, 25]. This situation does not mimic the dilemma of the radiologist and surgeon in clinical practice who are faced with a wide spectrum of possible diagnoses. Our series included 14 distinct pathologic diagnoses (Table 1), which more fairly reflects daily reality. In other published surgical series, the accuracy of preoperative diagnosis has been equally modest. For example, in a multicenter study of 164 serous cystadenomas and 222 mucinous tumors, the correct diagnosis was made preoperatively in just 20% of serous cystadenomas, 30% of mucinous cystadenomas, and 29% of mucinous cystadenocarcinomas [26]. In that series, the low rates of accurate diagnosis are undoubtedly partially a function of the multicenter nature of the study. However, even in experienced centers, serous cystadenoma is preoperatively differentiated from mucinous cystic tumor in fewer than 50% of cases overall [27–29].
Our finding that unilocular thin-walled cysts smaller than 4 cm were malignant in 13% (2/15) of cases is somewhat discrepant with the results of a recent study from Massachusetts General Hospital in which 35 of 36 unilocular pancreatic cysts smaller than 3 cm were benign [30]. The reason for this discordance is unclear but may partially reflect selection bias: 45 of the 58 patients in our series underwent surgical resection compared with just 48 of the 86 patients in the Massachusetts General Hospital series. It is also important to note, with respect to both our study and the Massachusetts General Hospital series, that the end point of frank malignancy is not sufficient for clinical management because even though entities such as mucinous cystadenomas are benign, they are considered to be premalignant. Although criteria for selective observation are emerging [31, 32], the current standard recommendation for the surgical care of suspected mucinous cystadenoma of the pancreas remains resection rather than observation in patients who are reasonable candidates for surgery [7].
Our study has a number of limitations. Unquestionably, the major limitation is the inclusion of only histopathologically proven cases. This criterion introduces a sizeable selection bias because truly classic-appearing benign lesions, such as a prototypical pseudocyst or serous cystadenoma (microcystic mass with central, stellate calcification), that did not undergo histopathologic verification were excluded. Conversely, because our institution is a large referral center, our series likely included some very rare entities that might never be encountered during the professional lifetime of a general community radiologist. An example of the latter is the case of a gastric duplication cyst within the pancreas mimicking a pseudocyst, a phenomenon that has been reported on only a few occasions [33]. Still, without pathologic confirmation or very long-term follow-up, it is difficult to claim diagnosis with confidence based exclusively on imaging, and it is the problematic masses that present for surgery that most require accurate characterization.
Another major limitation is that not all patients underwent both CT and MRI, with only 18 patients undergoing MRI and only 12 of those also undergoing CT. As such, it is likely that we are comparing heterogeneous populations when we compare the accuracy of both imaging techniques, because presumably clinical factors influenced the decision to refer patients for MRI, and these factors may have added additional selection biases that might have altered the accuracy of MRI in that subset. Furthermore, 17 patients who underwent CT did not have images obtained in the arterial phase; however, although arterial phase imaging may improve detection of solid pancreatic adenocarcinoma and help characterize hypervascular masses such as neuroendocrine tumor, we are unaware of any reports suggesting arterial phase images are of substantial benefit in the characterization of cystic pancreatic masses.
Our study was retrospective and undertaken at a single institution, which may have introduced unintended biases. Our study evaluated scans obtained over a 7-year period, and it is possible that imaging advances and technical heterogeneity over that period may have partially confounded our results. For example, postgadolinium imaging was performed with 2D rather than 3D spoiled gradient sequences during the study period, even though the latter is increasingly the standard of practice. Still, it is probably wishful thinking to believe that technologic advances will fundamentally change the likely fact that morphology is an imperfect predictor of pathology in cystic pancreatic lesions.
The criteria by which reviewers arrived at a given diagnosis were largely opaque because morphologic and demographic factors are inextricably linked in this evaluation, and our methodology did not allow us to tease the contribution of individual features that may have been at play in reviewer judgment. However, the high degree of interobserver agreement between both reviewers suggests that irrespective of which individual factors may have been at work, both radiologists arrived at similar conclusions. This level of agreement suggests that the similar but limited accuracy shown by both reviewers was not a function of poor reviewer judgment, but of fundamental limitations in the described features for characterization of cystic pancreatic masses.
The failure to show any useful diagnostic variables in our multivariate analysis may well be due to the inclusion of the entire spectrum of cystic neoplasms, with consequently only a small number of each specific subtype. For that reason, our analysis focused on the distinction of benign from malignant masses. Our decision to group premalignant lesions as benign could well be criticized because surgery is arguably the standard of care for lesions such as mucinous cystic neoplasms without frank carcinoma and intraductal papillary mucinous neoplasm with dysplasia. However, the reality is that surgical standards continue to evolve because our understanding of the natural history of many of these pathologic entities is incomplete; in an elderly patient with multiple comorbidities, it is only the presence of definite malignancy that is likely to lead to early surgical intervention.
Another potential criticism is that our hypothesis that small thin-walled unilocular lesions would likely be benign was speculative. However, it seemed a reasonable hypothesis to study not only because of the analogies with cystic lesions in the ovary and kidney but also because pseudocyst is a leading consideration for a benign unilocular lesion. A similar hypothesis was recently explored by investigators at the Massachusetts General Hospital [30]. Finally, we did not investigate the potential role of endoscopic sonography in the characterization of cystic pancreatic masses, although results from major referral centers have suggested this technique may be helpful [16].
In conclusion, CT and MRI are reasonably and similarly accurate in the characterization of cystic pancreatic masses as benign or malignant; limitations include a substantial rate of misdiagnosis even when reviewer certainty is high and a moderate frequency of malignancy in small morphologically benign-appearing cysts.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  N. Inan, A. Arslan, G. Akansel, Y. Anik, and A. Demirci Diffusion-Weighted Imaging in the Differential Diagnosis of Cystic Lesions of the Pancreas Am. J. Roentgenol., October 1, 2008; 191(4): 1115 - 1121. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 H. Ogawa, S. Itoh, M. Ikeda, K. Suzuki, and S. Naganawa Intraductal Papillary Mucinous Neoplasm of the Pancreas: Assessment of the Likelihood of Invasiveness with Multisection CT Radiology, September 1, 2008; 248(3): 876 - 886. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. S. Katz, D. M. Friedel, D. Kho, N. Georgiou, and J. J. Hines Relative Accuracy of CT and MRI for Characterization of Cystic Pancreatic Masses Am. J. Roentgenol., September 1, 2007; 189(3): 657 - 661. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
