| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, University of Yamanashi, Nakakoma, Japan.
2 Department of Radiology, Brigham and Women's Hospital, Harvard Medical School,
Boston, MA 02120.
3 Present address: Department of Radiology, Sisli Etfal Training and Research
Hospital, No: 10/1 Dogancilar, Uskudar Istanbul 81160, Turkey.
4 Department of Radiology, Yamanashi University, Shimokato, Japan.
Received June 16, 2005;
accepted after revision October 16, 2005.
Address correspondence to S. M. Erturk.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Thirty-one patients with and 35 patients without pancreatic adenocarcinoma underwent triple-phase MDCT. Three radiologists independently attempted to detect pancreatic adenocarcinoma and assess local extension using the MDCT images in five sessions. The first three sessions involved sets of images obtained in arterial phase, pancreatic parenchymal phase, and PVP separately and respectively. In the fourth session, a combination of axial images from all phases was evaluated. During the fifth session, radiologists had access to coronal and sagittal MPR images together with the axial images obtained in all phases. Results were compared with surgical findings using receiver operating characteristic (ROC) analysis and kappa statistics.
RESULTS. Regarding tumor detection, the image set composed of coronal and sagittal MPR images and of axial images obtained in all phases had a significantly higher value for the area under the ROC curve (Az, 0.98 ± 0.01) than the other image sets and yielded the highest sensitivity (93.5%). The sensitivity of the arterial phase image set (80.6%) was significantly lower than that of all other image sets. Whereas the image set composed of coronal and sagittal MPR images and axial images obtained in all phases yielded the highest kappa values for all local extension factors evaluated, the image set composed of only arterial phase images yielded the lowest kappa values for almost all of the factors.
CONCLUSION. A combination of pancreatic parenchymal phase and PVP imaging is necessary and efficient for the assessment of pancreatic adenocarcinoma. The addition of coronal and sagittal MPR images increased the performance of MDCT, especially in the evaluation of local extension.
Keywords: MDCT • oncologic imaging • pancreatic adenocarcinoma • pancreatic cancer
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CT is the established method for diagnosing and staging pancreatic adenocarcinoma [1, 4]. The introduction of MDCT has allowed further refinements in detecting pancreatic adenocarcinoma and in determining unresectability [1, 9-11]. Multidetector row technology allows acquisition of high-resolution images of the pancreas in all three phases of vascular en hancement—that is, the arterial phase, pancreatic parenchymal phase, and portal venous phase (PVP)—with the possibility that images may be reformatted and additional images in any plane can be used for assessment [9, 12].
Several scanning protocols have been described for imaging pancreatic adenocarcinomas [13-18]. However, only a few articles evaluate MDCT in this setting. Recently, McNulty et al. [9] and Fletcher et al. [10] reported that pancreatic parenchymal phase images and PVP images were equivalent to each other and were superior to arterial phase images regarding both tumor-to-pancreas attenuation difference and tumor conspicuity. Nevertheless, those studies did not include either a receiver operating characteristic (ROC) curve analysis or a comprehensive statistical evaluation for the detection of pancreatic adenocarcinoma and assessment of local extension. More recently, Imbriaco et al. [19] reported that a single acquisition between the pancreatic parenchymal phase and PVP with a caudocranial scanning direction is adequate for tumor detection and assessment of resectability; controversy about the use of multiplanar reformatted (MPR) images for detecting and assessing pancreatic adenocarcinoma still exists [4, 20, 21].
In this retrospective study, we aimed to evaluate the relative contributions of imaging phases and sagittal and coronal MPR images for the detection and preoperative local staging of pancreatic adenocarcinoma using a 16-MDCT scanner.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
During surgery, the presence of hepatic metastatic lesions was evaluated by means of intraoperative sonography. Regional lymph nodes were routinely sampled and were histopathologically evaluated during the operation. Surgeons also paid attention to the presence of peritoneal dissemination and ascites. In all cases, all factors for assessment of local tumor extension and vascular invasion were reevaluated histopathologically after surgery.
A control group of 35 patients with suspected pancreatic disease who had undergone the same triple-phase CT protocol as the study group and had negative results for pancreatic adenocarcinoma was identified from our clinical database. The control subjects had a clinical and CT follow-up of at least 12 months, during which there was no evidence of pancreatic adenocarcinoma. There were 19 women and 16 men in the control group. They ranged in age from 52 to 78 years (mean, 63 years), and their body weight ranged between 46 and 71 kg (mean, 58 kg). These 35 patients without adenocarcinoma underwent MDCT examinations to exclude pancreatic carcinoma that was suspected due to an elevation of cancer antigen 19-9 (n = 4), chronic pancreatitis (n = 18), intraductal papillary mucinous cystic tumors (n = 7), or acute pancreatitis (n = 6). The patients with acute pancreatitis were follow-up cases who were treated for acute pancreatitis within 3-6 months before the present study.
Our institutional review board approved this study, and informed consent was obtained from all subjects enrolled.
MDCT
All multiphasic contrast-enhanced MDCT examinations were performed on a
commercially available MDCT scanner (Aquilion, Toshiba Medical Systems) with a
gantry rotation speed of 0.5 second and a detector configuration of 16 x
0.5 mm; the moving speed of the table was 11 mm/s. All scans were acquired in
a cephalocaudal direction. Twenty minutes before the examination, patients
were given 500 mL of water for demarcation of the stomach and duodenum and
delineation of the pancreatic head region.
The unenhanced MDCT series were programmed to include the entire pancreas at first. The upper and lower MDCT scanning levels were chosen using these unenhanced CT images. The total scanning length for each phase was 7-10 cm. Patients underwent unenhanced MDCT followed by arterial phase, pancreatic parenchymal phase, and PVP imaging. All patients received 100 mL of IV iomeprol (Iomeron, 350 mg/mL, Bracco-Eisai; total dose of iodine, 35 g) with a monophasic injection technique by means of a power injector. The contrast material was administrated at a rate of 3 mL/s in all patients. Each phase acquisition was initiated 30 seconds for the arterial phase, 40 seconds for the pancreatic parenchymal phase, and 70 seconds for the PVP after the injection of contrast material began.
All MDCT examinations were performed at 120 kVp and 150 mAs. After the MDCT examinations, coronal and sagittal MPR images were also created using the data from pancreatic parenchymal phase images. For reconstructed axial and MPR images, a 5-mm slice thickness and 5-mm interval were used.
Image Analysis
For conducting ROC analysis, two radiologists serving as study coordinators
reviewed all MDCT images with knowledge of the clinical-pathologic findings.
Based on clinical and pathologic reports, they attempted to determine the
location of the lesions and to anatomically correlate the pathologically
confirmed lesions with the imaging findings as accurately as possible to allow
detection of false-positive readings. Therefore, for cases in which the
reviewers identified lesions other than the true lesions indicated by the
study coordinators, the study coordinators considered those lesions as
false-positive lesions.
Three experienced abdominal radiologists then interpreted the MDCT images of the patients with adenocarcinoma and of the control subjects independently using diagnostic monitors. The review process was performed in five separate sessions. The study coordinators determined the patient order for the review process. The images used for the first three sessions of the review process consisted of MDCT images obtained during individual enhancement phases (arterial phase, pancreatic parenchymal phase, PVP) in a randomized fashion for individual reviewers. The fourth session was appropriated for review of axial MDCT images from all phases, and the fifth session was allowed for the review of sagittal and coronal MPR images and the axial MDCT images from all phases. Thus, five different MDCT techniques—including arterial phase, pancreatic parenchymal phase, PVP, images from all phases, and images from all phases combined with MPR images—were evaluated in this study. The reviewers were aware that all the MDCT examinations were performed for evaluating suspected pancreatic carcinomas. However, they were blinded to all other information, such as patient identity, clinical history, and the results of other imaging examinations and histopathologic evaluations. Each review session was performed with intervals of 4 weeks to minimize learning bias.
Pancreatic carcinoma was defined as a hypoattenuating lesion with diminished enhancement relative to the normal peripheral pancreatic parenchyma on MDCT images in this study. For all MDCT images, each reviewer graded the presence (or absence) of a pancreatic carcinoma using the following 5-point confidence scale: 1 = definitely absent, 2 = probably absent, 3 = equivocal, 4 = probably present, 5 = definitely present. If a pancreatic carcinoma was considered to be present in the pancreas, the possible size and location of the tumor were recorded. The purpose of this study was not only to investigate tumor detection but also to investigate the usefulness of MDCT images for assessing local extension. Therefore, an assessment of local extension that included evaluation of local extension, vascular invasion, and lymph node involvement was also performed.
The following factors were evaluated regarding local extension of pancreatic adenocarcinoma: serosal and retroperitoneal peripancreatic extension and choledochal and duodenal invasion. Serosal and retroperitoneal extension were considered positive if the MDCT images revealed tumor contiguous to the adjacent organs, irregularly increasing attenuation of the peripancreatic fat layer, or spicular formations toward the peripancreatic fat layer on the pancreatic surface. Choledochal and duodenal invasion were considered positive if a suspected hypoattenuated lesion directly invaded or reached the surface of the common bile duct and duodenum, respectively, without evidence of intervening hyperattenuating normal pancreatic parenchyma. Local tumor extension was considered absent when evidence of intervening hyperattenuating normal pancreatic parenchyma between the tumor and the surface of the pancreas or the margin of the neighboring anatomic structure could be identified. Regional lymph nodes with a short-axis diameter of greater than 10 mm were considered positive for involvement [22].
Portal venous invasion (including the main portal vein, superior mesenteric vein, and splenic vein) and arterial invasion (including the celiac axis, common hepatic artery, and splenic artery) were evaluated as vascular invasion factors using the imaging criteria defined in previous studies [19]. A vessel was considered to be involved if it showed a focal reduction in caliber, circumferential (> 180°) encasement by tumor, or frank thrombosis. Portal venous invasion and arterial invasion were considered negative when the tumor was confined to the pancreas with normal perivascular fat planes.
Statistical Analysis
The interobserver agreement between reviewers for tumor detection, local
extension, and vascular invasion was calculated with the chance-corrected
kappa statistic. In general, a kappa statistic greater than 0.75 was
considered as excellent agreement beyond chance; 0.4-0.75, fair to good
agreement; and less than 0.4, poor agreement. Values near zero or less than
zero reflected only chance agreement
[23].
Composite ROC curves were used to represent the performance of the three reviewers for tumor detection as a group. These curves were calculated by averaging the binormal parameter values of the curves of individual reviewers. The findings were analyzed by means of the maximum likelihood estimation of a binormal ROC curve grading data [24]. The diagnostic accuracy of each MDCT technique for each reviewer was evaluated by calculating the area under the ROC curve (Az). The Az value combines true-positive, false-positive, false-negative, and false-positive ratings and level-of-confidence grades, and it represents a measure of the trade-off between detecting true-positive versus false-positive lesions (percentage of true-positive findings before the first false-positive finding) over a range of thresholds [25]. The mean Az values of individual MDCT techniques were calculated and compared using the jackknife method [26, 27]. Az values greater than 0.80 were considered to indicate good diagnostic accuracy, based on the findings of a previous study that used ROC analysis [24]. Regarding the ROC analysis, lesions assigned grade 4 or 5 (probably or definitely present, respectively) were considered positive for adenocarcinoma. The sensitivities of the individual MDCT techniques were calculated and subsequently were compared using a two-sided exact McNemar test [28].
The agreement between the results of the individual MDCT techniques and those of the surgical-pathologic results was evaluated using chance-corrected kappa statistics [29, 30]. This analysis was performed only for true-positive cases—that is, for cases having pathologically proven adenocarcinomas that were correctly identified by the reviewers.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Tumor Detection
The mean Az values and sensitivities of the individual
MDCT techniques for the detection of pancreatic adenocarcinoma are summarized
in Table 1. The mean
Az value of the combination of MDCT images from all phases
and MPR images was significantly higher than those of all other MDCT
techniques (p < 0.01). The mean sensitivity of the image set
composed of coronal and sagittal MPR images and axial images obtained in all
phases (93.5%, 81/87) was also the highest among all others and was
significantly higher than that of arterial phase images. For the mean
sensitivity, there were no significant differences among the combination of
MDCT images from all phases, pancreatic parenchymal phase images, PVP images,
and arterial phase images (p > 0.05). Figures
1A,
1B,
1C,
2A,
2B,
2C,
3A, and
3B show representative patients
with pancreatic adenocarcinoma.
|
|
|
|
|
|
|
|
|
Local Tumor Extension and Vascular Invasion
The kappa values indicating the agreement between surgical-pathologic
evaluation and image interpretation are shown in
Table 2. For all local tumor
extension and vascular invasion factors, the highest mean kappa values were
obtained with the image set composed of axial images obtained in all phases
and coronal and sagittal MPR images. For that image set, the mean kappa values
were in the category of excellent for choledochal invasion (
= 0.82),
lymph node involvement ( = 0.78), and portal venous invasion ( =
0.77); the mean kappa values were in the category of fair to good for serosal
extension ( = 0.66), arterial invasion ( = 0.69),
retroperitoneal extension ( = 0.56), and duodenal invasion ( =
0.51). No other MDCT technique showed a mean kappa value indicating excellent
agreement for any local extension and vascular invasion factor. For all
factors evaluated, the image set composed of axial images from all phases
showed the second highest mean kappa values of the study.
|
The mean kappa values found for pancreatic parenchymal phase images and PVP images were fair to good for all local tumor extension factors. Comparing the pancreatic parenchymal phase with the PVP, the pancreatic parenchymal phase had better agreement than the PVP for arterial invasion and lymph node involvement and vice versa for serosal and retroperitoneal extension and portal venous invasion. For duodenal and choledochal invasion, the mean kappa values found for the pancreatic parenchymal phase and PVP were equal to each other. All kappa values for the pancreatic parenchymal phase or PVP were higher than those found for the arterial phase.
Interobserver Agreement on Tumor Detection, Local Extension, and Vascular Invasion
The chance-corrected kappa () values that indicate interobserver
agreement on tumor detection among the three reviewers are shown in
Table 3. For all MDCT
techniques, the kappa values were fair to good ( = 0.4-0.75) between
reviewers 1 and 2. Between reviewers 2 and 3, the kappa values were excellent
( > 0.75) for arterial phase and PVP and were fair to good (
= 0.4-0.75) for pancreatic parenchymal phase, MDCT images obtained in all
phases, and MDCT images obtained in all phases and MPR images. There were no
kappa values indicating poor agreement for any reviewer pair.
|
Table 4 summarizes the kappa values indicating interobserver agreement for individual MDCT techniques for local extension and vascular invasion factors. For all MDCT techniques, the interobserver agreement was fair to good for all reviewer pairs.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Regarding the diagnostic performance of multiphasic CT in the detection of pancreatic adenocarcinomas, Choi et al. [17] reported that tumor detectability on arterial phase images obtained from 30 seconds after the administration of contrast material (95%) was superior to that on PVP images obtained from 80 seconds after administration of contrast material (68%). Conversely, Graf et al. [15] and Keogan et al. [14] reported that the addition of arterial phase images to PVP images did not contribute to an improvement in the detection of pancreatic adenocarcinomas. Meanwhile, Lu et al. [18] introduced a new concept: pancreatic parenchymal phase imaging. They suggested that images be obtained beginning 40 seconds after the administration of contrast material—during the so-called pancreatic parenchymal phase; according to Lu et al., images obtained during this phase showed the maximal tumor-to-pancreas contrast. However, the findings of all these previous reports depend on the results of studies performed on single-detector CT scanners.
Recently, few investigators have attempted to define an optimal MDCT protocol for the evaluation of pancreatic adenocarcinomas; to our knowledge, only two studies have compared different contrast-enhancement phases including arterial phase, pancreatic parenchymal phase, and PVP using MDCT scanners. McNulty et al. [9] mentioned that pancreatic parenchymal phase and PVP were equivalent to each other and superior to arterial phase regarding both tumor-to-pancreas attenuation difference and tumor conspicuity. Nevertheless, in our opinion their analysis was incomplete because sensitivities of three such different phases for tumor detection were not calculated. Fletcher et al. [10] made the same conclusion as McNulty et al.; they mentioned that the sensitivities for tumor detection of pancreatic parenchymal phase (97%, 29/30) and PVP (93%, 28/30) imaging were superior to those achieved with arterial phase imaging (63%, 19/30). Nevertheless, their statistical calculations did not include a comprehensive ROC analysis and, therefore, are not strong enough to end the controversy about an optimal MDCT protocol for the evaluation of pancreatic adenocarcinomas.
Most recently, Imbriaco et al. [19] reported that single-phase MDCT is an accurate technique for the diagnosis and assessment of resectability in patients with a suspected pancreatic neoplasm. In that study, the authors used a scanning delay of 60 seconds and patients were scanned in the caudocranial direction. However, the authors also included pancreatic cancers other than adenocarcinomas in their study. In particular, cystic tumors were included in the pancreatic cancer group. This might have contributed to the high sensitivities of the two reviewers (97.3% and 95.8%) for tumor detection. Furthermore, Imbriaco et al. analyzed resectability globally and did not evaluate individual local extension factors, and only nine of their patients had surgically resectable disease.
Our results based on ROC analysis are similar to those of McNulty et al. [9] and Fletcher et al. [10]. In our study, Az values indicating comprehensive diagnostic performance of the different MDCT techniques and the sensitivity values for tumor detection with pancreatic parenchymal phase and PVP were identical (Az = 0.95, sensitivity = 89% [83/93]); moreover, they were virtually equal to the sensitivity and Az value of the image set composed of MDCT images from all phases (Az = 0.96, sensitivity = 90% [84/93]).
The kappa values indicating the level of agreement between interpretations of the pancreatic parenchymal phase and PVP image sets and the surgical-pathologic results for each local staging factor might be considered to complement one another. The kappa values were superior with the pancreatic parenchymal phase for arterial invasion and lymph node involvement and with the PVP for serosal and retroperitoneal extension and portal venous invasion. In addition, a combination of kappa values for pancreatic parenchymal phase and PVP imaging was approximately equal to those of the image set composed of MDCT images obtained in all phases. Thus, our results showed that both pancreatic parenchymal phase and PVP imaging should be included in an MDCT protocol for the assessment of local extension of pancreatic adenocarcinomas. A combination of pancreatic parenchymal phase and PVP images was necessary for MDCT detection and assessment of pancreatic adenocarcinomas; there was no need to include images obtained in the arterial phase in an MDCT protocol in this context.
One of the most valuable advantages of MDCT scanners is their ability to generate MPR images or 3D volume-rendered images. MPR images obtained with MDCT images can realize high image quality with high resolution without slice misregistration because the row data generated by MDCT images are obtained as a continuous volume with an isotropic voxel size [31].
To our knowledge, only one study has compared axial multiphasic MDCT images and curved MPR images for the detection and local staging of pancreatic adenocarcinomas [4], and a few review articles refer to the value of curved MPR images for the evaluation of pancreatic adenocarcinomas [4, 20]. Prokesch et al. [4] compared axial multiphasic MDCT image sets alone with the curved MPR image sets alone and concluded that curved MPR image sets and axial multiphasic MDCT image sets had equivalent abilities in the detection of pancreatic adenocarcinomas and in the determination of tumor resectability.
In the present study, we used two orthogonal MPR images rather than curved MPR images. Whereas curved MPR images are operator-dependent and time consuming to obtain [31], MPR images in the sagittal and coronal planes are easily reproducible. Furthermore, unlike the curved MPR images, the anatomy of the depicted structures generally remains comprehensible on the orthogonal plane MDCT images [31]. Our results show the benefits of adding coronal and sagittal MPR images to the study protocol; with the addition of coronal and sagittal MPR images, reviewers achieved an obvious improvement in both the detection and the local staging of pancreatic adenocarcinomas.
For tumor detection, the Az value with the combination of the axial MDCT and MPR images was significantly higher than those with all other MDCT techniques without MPR images. Again, kappa values indicating the ability of different techniques for assessing local staging were also significantly higher for all local staging factors with the combination of axial MDCT and MPR images than those obtained with any other MDCT technique. We believe that this improvement in the assessment of local staging with the use of a combination of axial MDCT and MPR images was mainly due to the 3D extension characteristics of pancreatic adenocarcinomas.
There are several limitations in our study. First, the study population of patients with pancreatic adenocarcinomas is small. Another limitation is that the present findings cannot indicate the absolute sensitivity and Az value of each MDCT technique for the prospective diagnosis of pancreatic adenocarcinomas because this study included only limited types of pancreatic diseases as the negative cases. If more subjects with other pathologic conditions were intermixed in this study, there would have been a more real-life situation. Furthermore, we might have compared the pancreatic parenchymal phase and PVP regarding their performances for the detection of hepatic metastases. Nevertheless, our results show that for the evaluation of local extension, these two techniques were complementary. Considering the other limitations, further studies are needed to confirm our results.
In conclusion, a combination of pancreatic parenchymal phase and PVP imaging is essential and is sufficient for an optimal multiphasic MDCT protocol for the comprehensive evaluation of pancreatic adenocarcinomas. The addition of coronal and sagittal MPR images to the MDCT protocol increases the sensitivity of MDCT and improves its agreement with surgical findings regarding local staging factors.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
