| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Seoul National University College of Medicine, Seoul
National University Hospital, 28, Yongon-dong, Chongno-gu, Seoul, 110-744
Korea.
2 Institute of Radiation Medicine, Seoul National University Hospital, Seoul,
Korea.
3 Department of Radiology, Konkuk University Hospital, Seoul, Korea.
4 Department of Radiology, Chung Ang University Hospital, Seoul, Korea.
Received June 17, 2006;
accepted after revision November 8, 2006.
Address correspondence to J. M. Lee.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. Institutional review board approval was obtained, but informed consent was not required for this retrospective study. Sixty-seven patients underwent dynamic multiphasic CT within 3 months before liver transplantation. Interval reviews using tile and multisynchronized stack modes were performed independently by four reviewers with various levels of experience to determine the presence of HCC using a five-point confidence scale. Observer performance was compared using jackknife free-response receiver operating characteristic (ROC) analysis. The time required to interpret the CT scans using each mode was recorded and compared using the paired Student's t test.
RESULTS. Twenty-seven patients had 48 HCC nodules. The mean
free-response ROC figures of merit for detecting HCC were significantly higher
using the multisynchronized stack mode (0.731) than using the tile mode
(0.662) (F-statistic = 6.603, p = 0.012). The 95% CIs for the task
were -0.125 - -0.016. The time used for image analysis was also significantly
shorter with the stack mode (63 
75 seconds) than with the tile mode (94
191 seconds) for all four reviewers (p < 0.0001).
CONCLUSION. Multisynchronized stack viewing of multiphasic dynamic CT scans significantly increases the detection rate of HCC in liver transplant candidates. It also significantly shortens the interpretation time compared with tile viewing.
Keywords: CT • diagnostic radiology • liver neoplasms • observer performance • PACS
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The next stage in the interpretation process is an advanced iteration of the stacked mode, called "multisynchronized stack mode," which allows synchronization of two or more individual stacks [1, 2]. This can take the form of linking two historical comparison studies or two or more individual phases or sequences within a single examination. Synchronizing multiple phases on CT or MRI allows the radiologist to correlate comparable images of the various phases being viewed. The advantage of multisynchronized stack mode for this purpose is even more apparent when viewing more complicated cases of liver imaging, which require multiple phases in one examination as described.
With the advent of the MDCT scanner, which allowed a significant improvement in the volume coverage speed, multiphasic CT images, including arterial (sometimes double arterial), portal venous, and delayed phase images, can be obtained for evaluating patients with hepatic tumors, in particular hepatocellular carcinoma (HCC) [3-6]. Furthermore, as several authors have emphasized, to take advantage of unenhanced or delayed phase images in the detection of additional tumors and marginal recurrence after transarterial chemoembolization for HCC [7, 8], three or four CT data sets of different phases are routinely obtained in many institutions for any patient with suspected HCC. From this perspective, side-by-side comparison of different enhancement patterns over multiple phases is crucial for HCC evaluation.
However, to our knowledge, the superior advantages in time efficacy of using the multisynchronized stack mode rather than the tile mode have been largely undocumented, especially in multiphasic hepatic imaging. Therefore, the purpose of our study is to compare and show the advantages of CT image interpretation with PACS multisynchronized stack mode over tile mode in terms of speed and observer performance for the detection of HCC in liver transplant candidates using a jackknife free-response receiver operating characteristic (ROC) analysis.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Our clinical findings for these patients are summarized in Table 1. The most common cause of cirrhosis was hepatitis B (54/67, 80.6%), and 46 (68.7%) of 67 patients were Child-Pugh class C. Thirteen patients received adjuvant treatment while waiting for a donor.
|
MDCT Technique
Most of the CT examinations were performed on either of two MDCT scanners,
Sensation 16 (Siemens Medical Solutions [n = 29]) or LightSpeed Ultra
(GE Healthcare [n = 29]). For the remaining nine patients, a 4-MDCT
scanner (MX 8000, Marconi Medical Systems [n = 3]) and a
single-detector CT scanner (Somatom Plus 4, Siemens Medical Solutions
[n = 6]) were used. The respective scanning parameters used for the
4-, 8-, and 16-MDCT scanners were detector configuration, 4 x 2.5, 8
x 1.25, and 16 x 0.75 mm; slice thickness, 3.2, 2.5, and 3 mm;
reconstruction interval, 3, 2.5, and 3 mm; table speed, 12.5, 13.5, and 24
mm/rotation; and 150, 250, and 200 effective mAs with rotation time, 0.5
second; 120 kVp; and matrix, 512 x 512. For single-detector CT, a 5-mm
slice thickness and reconstruction interval, table feed of 7 mm/rotation,
rotation time of 1 second, 165 mAs, and 120 kVp were used.
With 8- and 16-MDCT scanners, triple-phase CT (i.e., unenhanced, early hepatic arterial, late hepatic arterial, and portal venous phases) was performed. At first, a baseline unenhanced scan was obtained through the entire liver. Patients received 120 mL of nonionic contrast material with a concentration of 370 mg I/mL IV using a power injector at a rate of 3 to 4 mL/s. The scanning delay for the early hepatic arterial phase was 5 seconds after reaching the enhancement of the descending aorta up to 100 H as measured by a bolus-tracking technique [9]. For the late hepatic arterial and portal venous phases, the interscan delays were 9 and 30 seconds for a 16-MDCT scanner and 7 and 20 seconds for an 8-MDCT scanner, respectively. Therefore, each dynamic scanning session began after approximately 23, 40, and 78 seconds for a 16-MDCT scanner and after 23, 44, and 78 seconds for an 8-MDCT scanner after the initiation of the contrast injection. For a 4-MDCT and a single-detector CT scanner, triple-phase CT (i.e., unenhanced, arterial, and portal venous phases) was performed. The scanning delay was 11 and 45 seconds for the arterial and portal venous phases, respectively, after reaching the enhancement of the descending aorta up to 100 H as measured using a bolus-tracking technique.
Image Analysis
All 67 CT scans were analyzed independently and retrospectively by four
reviewers who were blinded to the presence of HCC and any clinical
information, including the 
-fetoprotein level, other than the
designation, "liver transplant candidates." The four reviewers
included two faculty radiologists with 11 and 9 years of experience, one
abdominal fellow with 7 years of training, and one senior resident with 3
years of training. All reviewers were well acquainted with the use of PACS.
The two faculty radiologists had experience interpreting hard-copy CT images
for 6 and 3 years, respectively, whereas the fellow and resident did not have
hard-copy interpreting experience.
The four observers interpreted the CT images using the two interpretation modes at a 2-month interval. Evaluation using the multisynchronized stack mode preceded evaluation using the tile mode. A PACS workstation with three 2,048 x 1,536 pixel, 20.8-inch (52.8 cm) monochrome liquid crystal display (LCD) monitors was used for both stack and tile mode evaluation. The reviewers could freely control both the window width and the level settings. In both modes, each observer was allowed to choose the number of images displayed per screen, but usually nine or 12 for the tile mode and four for the stack mode were recommended.
All observers were asked to record the number and sizes of the HCCs in each of the eight hepatic segments and to assign a confidence level for the diagnosis of HCC for the jackknife free-response ROC analysis. Free-response ROC data consists of mark-rating pairs in which the number of marks per image is determined by the reviewer and could be as small as zero [10]. Lesion size was estimated by measuring the maximum diameter on the transverse CT images with an electronic ruler. Diagnostic confidence for each lesion was subjectively scored on a five-point scale (0, no HCC; 1, probably absent; 2, possibly present; 3, probably present; and 4, definitely present). Before interpreting the images, the reviewers were informed that the categorization of confidence levels of 2 or higher represented a positive diagnosis of HCC. All lesions assigned a confidence level of 2 or higher and confirmed to be HCC were considered to be true-positive diagnoses. All lesions assigned a confidence level of zero or 1 when a lesion was actually proven to be HCC were considered false-negative diagnoses.
For objectivity and reproducibility of the image analysis performed in this study, the criteria for HCC were provided. On CT images, nodules showing enhancement during the hepatic arterial phase and a lack of portal venous supply were regarded as HCC nodules [11-14]. In addition, a hypovascular and hypoattenuating lesion of 1 cm or larger in size that did not meet the criteria for a cyst or confluent fibrosis was also considered to represent HCC [15, 16]. The criteria for a cyst are a sharply delineated round or oval lesion with attenuation near that of water and no contrast enhancement of the wall or contents, and the criteria for confluent fibrosis are a focal, hypoattenuating, wedge-shaped lesion radiating from the portal fissure and associated with parenchymal atrophy with overlying capsular retraction and lack of displacement of vessels [15].
In addition, the numbers of false-positives and false-negatives for each interpretation mode for each reviewer were counted by two other radiologists. False-positive lesions were classified into two types according to their enhancement characteristics: hypoattenuating and hyperattenuating. The reviewers further analyzed the causes for false-negatives at a retrospective unblinded review. The time required to assess the CT images was also measured with a stopwatch by each observer.
Reference Standard
When we collected the study population, the MR images and the CT images
were interpreted by consensus between two other gastrointestinal radiologists
who did not participate in the CT interpretation sessions for the pathologic
evaluation. All lesions suspected of being HCC nodules were identified, and
the pathologist was notified. Gross and histologic analyses of all explanted
livers were performed by an experienced hepatobiliary pathologist. The
presence or absence of all lesions identified with one or more of the imaging
tests—that is, CT or MRI—was determined histologically on a
lesion-by-lesion basis.
In our study, we used MR images only to strengthen the accuracy of tumor detection by the pathologist when the explants correlated with the images. In addition, because MRI findings are not the purpose of this investigation, we did not analyze or correlate them with CT findings. All explanted livers were initially sectioned at 5-mm or thinner intervals in a sagittal plane. If a lesion identified on an imaging test could not be shown in the explant liver, representative histologic sections were obtained from the region of the liver that best corresponded to the lesion seen on the imaging test. In those patients in whom a lesion was not seen on an imaging test, the pathologic specimens were carefully reviewed for the presence of HCC.
Statistical Analysis
For imaging at each interpreting session, a jackknife free-response ROC
analysis method [10] was
performed using the jackknife free-response ROC analysis software that is
available at the Dev Chakraborty Website
[17]. For each image
interpretation yielding varying numbers of mark-rating pairs, a localization
criterion was applied to classify each mark as true-positive or
false-positive, and the figure of merit (
) was defined as the
probability that a true-positive rating exceeds a false-positive rating in a
normal image [10]. An
F-statistic test was used internally for the analysis of variance, yielding a
p value for rejecting the null hypothesis of no difference between
the techniques. In addition, the numbers of false-positives and
false-negatives between the two interpretation modes for each reviewer were
compared using the paired Student's t test. The differences in
interpretation times were also assessed using the paired Student's t
test. A p value of 0.05 or smaller was considered to indicate a
statistically significant difference.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Observer Performance of Multiphasic Liver CT Using Two Interpretation Modes
The reviewer-averaged values of the free-response ROC figure of merit
() for the tile and multisynchronized stack modes were 0.662 and 0.731,
respectively (Table 2). In
addition, the 95% confidence interval for 
(the difference in
the figures of merit between two interpretation modes) was -0.125 -
-0.016. The jackknife free-response ROC analysis indicated that the difference
between the two interpretation modes was significant at the 5% level for the
detection of HCC (F-statistic = 6.603, p = 0.012).
|
The total number of false-positives detected by the four reviewers was 136 for the tile mode and 102 for the stack mode, and the total number of false-negatives detected by the four reviewers was 94 for the tile mode and 83 for the stack mode. Details of the false-positives and false-negatives of each reviewer in each interpretation mode are presented in Table 3. The average numbers of false-positives and false-negatives derived from the four reviewers for each interpretation mode are listed in Table 4 along with the p value. The average numbers of false-positives tended to be smaller in the stack mode than in the tile mode for all reviewers except one of the faculty, but the differences were statistically significant only for the fellow (p = 0.042) and the resident (p =0.005). The average numbers of false-negatives also tended to be smaller in the stack mode than in the tile mode; however, the differences between the two modes failed to obtain statistical significance for all four reviewers.
|
|
Interpretation Time
All four reviewers evaluated multiphasic dynamic CT significantly faster
with the multisynchronized stack mode than with the tile mode (p <
0.0001) (Table 5), with the
improvement in speed varying by a factor of 1.3 to 2.7.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Although the liver is not a sophisticated organ such as the small bowel in which radiologists must scroll backward and forward to reveal longitudinal relationships, it needs at least three different CT phases—that is, unenhanced, arterial, and portal venous—to determine the enhancement pattern of a lesion. Such multiphasic CT techniques inevitably increase the number of images and enhance the complexity and difficulty of evaluation. As in the stack mode, a single monitor can hold four stacked CT images of four different phases simultaneously for comparison of the enhancement pattern on each phase, and comparison images are close together rather than up to several feet away as can be the case with the tile mode [1, 2]. Therefore, radiologists can compare CT images of three or four phases with minimal movement of their eyes from a given location on the monitor. The tile mode, on the other hand, requires observers to constantly shift their gaze between multiple images displayed in separate spatial fields. Particularly with large data sets of images, this may lead to a longer interpretation time.
More important, the diagnostic performance in terms of the average
value obtained using the stack mode evaluation was significantly higher than
that using the tile mode. Such results may be attributed to several factors.
First, in our study, the total number of false-positives and that of
hyperattenuating false-positive lesions in the stack mode were smaller than
those in the tile mode. Such a tendency was statistically proven to be
significant for the fellow and the resident who did not have sufficient
experience using the tile mode.
It is known that the most common cause of hyperattenuating false-positive interpretations in the evaluation of HCC is nontumorous arterioportal shunt [19]. The dynamic CT findings of arterioportal shunt have been summarized as wedge-shaped transient parenchymal enhancement during the arterial phase and early enhancement of peripheral portal veins, which manifest as hyperattenuating linear branching structures before the central portal vein is enhanced [20-22]. However, in actual clinical settings, we frequently find it difficult to differentiate between an arterioportal shunt and a small HCC for the following two reasons. First, some arterioportal shunts located in a longitudinal direction and manifested as a cone-shaped lesion may appear as nodular enhancing lesions in several contiguous transverse images. Second, when a small, brightly enhancing spot is seen within the enhancing lesion, it is not always easy to determine whether this dot is a peripheral portal branch or tumor neovascularity. For these reasons, we believe that the stack mode has some advantages over the tile mode.
With the tile mode, this cone-shaped arterioportal shunt lesion could be misinterpreted as a true enhancing nodule because the longitudinal information becomes fragmented by being displayed side by side. Furthermore, it may be difficult for reviewers to trace tiny peripheral portal branches connected to each other in several contiguous transverse images on the tile mode. On the other hand, with the stack mode, reviewers can more clearly navigate longitudinal relationships between contiguous images so that the observer can trace tubular structures, such as peripheral portal branches, with ease and confidence. Furthermore, because the stack mode improves the integration of information between adjacent images, thereby facilitating understanding of the 3D structure as a whole [23-25], radiologists can interpret multiple, small, enhancing nodular lesions as cone-shaped arterioportal shunts located in a longitudinal direction.
Second, the stack mode can considerably reduce the number of false-negatives. Indeed, in our study, there were fewer false-negatives caused by perception error in the stack mode (25) than in the tile mode (35). This means that radiologists can detect subtle lesions better with the stack mode than with the tile mode, and such a tendency was proven in our study even though it failed to obtain statistical significance.
Several factors may be responsible for fewer perception errors. The use of a larger image size in the stack mode allows radiologists to find subtle or small lesions more easily than in the tile mode. In addition, using the stack mode, radiologists can easily concentrate on a certain structure such as vessels or central or peripheral portions of the liver during the motion of scrolling up and down. From this action, radiologists can easily differentiate true enhancing nodules from the vessel that otherwise might have been misinterpreted as a normal vessel on static images. Furthermore, in the human retina, the amacrine and ganglion cells are more sensitive to change in incident light intensity than to the absolute value of incident light intensity [26]. Compared with interpreting multiple static images in the tile mode, scrolling through images in the stack mode results in moving interfaces that are theoretically more readily appreciated by the human visual system. Therefore, radiologists may detect subtle enhancing or washout lesions better with the stack mode than with the tile mode in the evaluation of dynamic liver CT.
There are several limitations to our study. First, all of our study patients who were candidates for liver transplantation and who had advanced liver cirrhosis were highly selected cases and therefore may not be representative of the general population encountered in routine daily practice. Therefore, it is likely that we may have overestimated the difficulty of evaluating HCC. Second, the image display size was larger in the stack mode than in the tile mode. Interpretation with larger CT images for the detection of HCC may lead to improvement of observer performance compared with that with smaller image size. However, this study was conducted in a clinical setting, and in such circumstances the images in the stack mode are larger than those in the tile mode. Viewing images in the tile mode at a size equal to that in the stack mode would be impractical because it requires radiologists to spend much more time to interpret CT data sets.
Third, there was a potential for reviewer bias because the observers themselves were asked to measure the time taken for interpretation (although a stopwatch was used). Also, because all observers performed the stack mode evaluation first, they might have been more accustomed to the images at the time of tile mode evaluation even if they did not recall the results of the previous evaluation. Fourth, even though there have been several reports on the usefulness of delayed CT images to improve detectability of HCC [5, 27], we did not use delayed phase as a routine CT protocol for liver cirrhosis because of practical and radiation dose concerns. This may have led to low detection performance of radiologists for HCC. Finally, we did not analyze in detail the causes of the false-positive lesions. Because our study was conducted in a retrospective manner, a direct comparison of the imaging and pathologic findings and histologic confirmations of false-positive lesions was not performed. This also accounts for why the results of our study regarding false-positives are different from those of a previous study in which most false-positives (20/35, 57.1%) were hypoattenuating [19].
In conclusion, multisynchronized stack mode evaluation clearly showed better observer performance and time efficacy than tile mode for the identification of HCC in liver transplant candidates on multiphasic dynamic CT. This was primarily due to the sheer number of screens that had to be examined on tile mode and the more efficient, intuitive process of having only one screen to look at in the multisynchronized stack mode, especially in the more complicated setting of liver imaging, which requires multiple phases in one examination. The use of stack viewing will aid radiologists to more easily perform the tasks of detecting and characterizing enhancing hepatic lesions in the cirrhotic liver.
Acknowledgments
We acknowledge Joon Koo Han and Jae Young Lee for their advice throughout
the preparation of the manuscript and during manuscript revision.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
