| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Mayo Clinic, 200 First St. SW, Mayo E-2, Rochester,
MN 55905.
2 Siemens Medical Solutions, Malvern, PA.
Received August 31, 2006;
accepted after revision October 11, 2006.
Address correspondence to J. G Fletcher
(fletcher.joel{at}mayo.edu).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. A colon phantom was scanned at 12, 25, 50, and 100 mA with standard CT colonographic acquisition parameters. Four reviewers using manual 2D methods and an automated polyp measurement tool measured 24 polyps of varying sizes and morphologic types, some at a haustral fold tip and some not at a fold tip. The accuracy (difference from true value) of manual and automated methods was compared across polyp sizes, morphologic types, locations, and doses. Precision (closeness of different measures) was compared for intraobserver and interobserver measurements.
RESULTS. The accuracy of automated polyp measurement was dependent
on morphologic type (p 
0.02), size (for three of four reviewers,
p 0.05), and location of polyps with respect to haustral folds
(two of four reviewers, p 0.01). For two of four reviewers,
automated measures were less accurate for 5-mm polyps, flat polyps, and polyps
at the tips of folds (p 0.04). Intraobserver precision was high,
two automated measurements being within 0.1 mm of each other 82-93% of the
time. Interobserver precision values for automated measures were more similar
85% of the time (82/96; p <0.001).
CONCLUSION. Accuracy of automated polyp measurements depends on polyp size, morphologic type, and location. When using an automated tool, radiologists should visually inspect automated polyp measurements, particularly for small and flat polyps and those located on folds, because manual measurements may be more accurate in this setting. Automated polyp measurements are more precise than manual measurements.
Keywords: colon cancer • CT colonography • polyps
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1%) or
high-grade dysplasia (3-5%) [9,
10]. In our colonographic practice, an increasing number of patients with polyps are undergoing follow-up with CT colonography, particularly when the polyps detected on colonography are of subcentimeter size or when patients have comorbid conditions that increase the risk of endoscopic polypectomy. The Working Group on Virtual Colonoscopy [11] has recommended that "for patients who do not have increased risk factors for development of colorectal carcinoma...it is reasonable to recommend interval surveillance when one or two 6-9 mm lesions are detected." Accurate measurement of polyp size is needed to assess the risk of malignancy and the need for referral for endoscopic polypectomy. Precise measurement of polyp size is needed to assess for interval growth of polyps on follow-up imaging. The purpose of our study was to assess the accuracy and precision of an automated polyp measurement tool on CT colonography of a colon phantom containing polyps of known sizes.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The colon phantom was submerged in a water bath simulating the attenuation of the body cavity and was scanned with a 64-MDCT system (Sensation 64, Siemens Medical Solutions) with four tube current values (12, 25, 50, and 100 mA). All other CT acquisition parameters were identical, corresponding to those used in clinical practice (effective tube current, 120 kVp; tube rotation time, 0.5 second; reconstruction kernel, B30f). All data sets were reconstructed axially with 1.0-mm slice thickness and 0.8-mm reconstruction interval.
CT colonography data sets were interpreted by reviewers using a Leonardo workstation and Syngo Colonography software package with an automated polyp measurement tool (Siemens Medical Solutions). Reviewers had 1-8 years of experience in interpreting CT colonographic data sets and had participated in one human or phantom study requiring manual polyp measurement. Each CT data set was evaluated by four reviewers, who reported linear measurements for 24 polyps. There were six polyps of each size (5, 7, 10, and 12 mm). The polyps in each size group were of three morphologic types (pedunculated, sessile, or flat) and two locations (located at the tip of a haustral fold or not located at the tip of a fold [i.e., between folds]).
Each reviewer evaluated four CT data sets (one at each dose level), performing two automated measurements and one manual measurement. We performed only one manual measurement because we had previously studied the accuracy and precision of manual measurements [13]. Reviewers performed the automated measurements by activating the automated measurement tool and clicking the computer mouse when the cursor was at various points along the surface of a polyp on the 3D endoluminal view. The second automated measurement was performed immediately after the first, but reviewers were instructed to rotate the 3D view and click on a different point on the polyp surface for the second measurement. The automated polyp measurement tool showed the size of the polyp and circumscribed the measured polyp on 2D and 3D views (Fig. 1A, 1B, 1C, 1D, 1E, 1F, 1G, 1H). Manual polyp measurements were performed with a linear line tool and oblique 2D multiplanar reformatted images at lung window settings [13]. Care was taken to measure the single largest linear dimension of the polyp head without the stalk [11].
|
|
|
|
|
|
|
|
|
Intraobserver agreement of automated measurement was estimated for each of the four observers by calculation of the difference between the two automated measurements. Interobserver agreement was estimated between reviewers 3 and 4, who evaluated the same data sets, by calculation of the difference between their first automated measurements and the manual measurements. Both interobserver and intraobserver agreement were estimated with the intraclass correlation coefficient. Each observer evaluated all polyps, and the reviewers in this study were assumed to be a random subset of radiologists.
Precision, both within an observer and between two observers, was estimated with the coefficient of variation and estimation of the number of times two different automated measures were 0.1 mm or less apart. A coefficient of variation is the relative variance of the measured values from the mean (i.e., SD of differences divided by mean value of the differences). For reviewers 3 and 4, who measured the same polyps in every data set, we also counted the number of times automated measurements between reviewers were closer than manual measurements between reviewers.
The precision of automated measurement compared with the precision of manual measurement was estimated, with precision being the absolute value of the difference between the automated or manual measurements. For reviewers 3 and 4, the association between this difference in precision of the automated and manual methods and the four polyp characteristics was made with linear regression, as was done for the accuracy of automated measurement.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
Polyp size—For one of four reviewers, there was a significant overall negative association between increasing polyp size and absolute error with the automated measurement tool (p = 0.04). For three of the four reviewers, the mean absolute error associated with automated measurement of 5-mm polyps was significantly greater than the mean absolute error associated with measurement of 12-mm polyps (p = 0.03-0.05). In comparing automated with manual measurement accuracy, two of four reviewers found that manual measurements had significantly better measurement accuracy for 5-mm polyps (p = 0.037) compared with 12-mm polyps (p = 0.05). No reviewer had a significant difference in accuracy between manual and automated measurements for polyps measuring 7 or 10 mm compared with 12-mm polyps.
Morphologic type—There was a significant association between
morphologic type and absolute error of automated measurement for each of the
four reviewers (p 0.02 for each). All four reviewers found that
the automated polyp measurement tool was more accurate in measurement of
pedunculated polyps compared with flat (p 0.01 for each of the
four) and sessile (p 0.05 for each) polyps
(Table 1). Two of four
reviewers found that manual technique improved the accuracy of measurement of
flat polyps (p = 0.023) compared with that of pedunculated polyps
(p = 0.05). There was no significant difference in accuracy between
manual and automated measurements of sessile compared with pedunculated polyps
(p >0.05 for each of these comparisons).
Polyp location—Two of four reviewers found significantly better accuracy of automated measurement for polyps not located on the tip of a fold (p = 0.001 and p < 0.01). Data from these reviewers indicated that polyps on the tips of folds, compared with those not on the tips, were measured more accurately with manual techniques (p = 0.001, p = 0.03). For the other two reviewers, accuracy of the automated tool was not significantly related to polyp position (p > 0.05).
Dose—Data from reviewers 1 and 2 showed that the dose setting during CT colonographic acquisition was not significantly associated with inaccuracy (p = 0.57-0.89). There was no significant association between dose level and accuracy as measured by the difference between manual and automated measurements for any dose level (p =0.36 to p =0.96).
Combining factors—Figure 3A, 3B shows the mean absolute error for different morphologic types of polyps located on the tip of a fold and not on the tip of a fold. When several factors influencing accuracy are combined, the absolute error can be large. For example, the mean absolute error for 5-mm flat polyps on the tip of a fold was 3.5 mm with the automated polyp measurement tool, compared with 0.8 mm with manual technique. No other combination of polyp factors resulted in a mean absolute error in the automated measurements greater than 2 mm. Figure 4A, 4B, 4C, 4D shows how large errors can arise when automated tools are used to measure flat or sessile polyps located on the tip of a fold. The automated tool can circumscribe a portion of a fold during measurement of the polyp.
|
|
|
|
|
|
Multivariable linear regression analysis showed that morphologic type was independently associated with inaccuracy of automated polyp measurement for all four reviewers (p = 0.0002-0.024). Small polyp size (5 mm) was independently and significantly associated with inaccuracy in automated polyp measurements for three reviewers (p = 0.01-0.03).
Precision of Automated Polyp Measurements
The average differences between the two automated measurements for each
polyp (across all polyp types and doses) for the four reviewers were 0.05
± 0.1, 0.03 ± 0.09, 0.05 ± 0.20, and 0.07 ± 0.2
mm, respectively. In most cases, there were no differences between the two
automated measurements performed by a single observer (range, 77-82% of
measurements). The number of times the two automated measurements were within
0.1 mm or less of each other were 83/96 (86%), 84/96 (88%), 79/96 (82%), and
89/96 (93%) for reviewers 1-4, respectively.
Table 3 shows the coefficients of variation representing the precision between automated and manual measurement for intraobserver and interobserver measurements. For all polyp characteristics, except 5-mm size and the interobserver agreement at 50 mA, the coefficients of variation were equal to or larger for manual measurements between reviewers than for automated measurements between reviewers.
|
Figure 5 shows the precision of automated measurements between reviewers 3 and 4 versus manual measurements for the same pair of reviewers. Eighty-five percent of the time (82 of 96 instances), the difference between the two automated measurements was smaller than the difference between the two manual measurements (95% CI, 77-92%; p < 0.001). Five percent of the time (five of 96 instances) the manual measurements were closer than the two automated measurements (95% CI, 2-12%). The mean ± SD and median for the differences between the two automated measurements were 0.07 ± 0.36 mm and 0.0 mm compared with 0.53 ± 0.43 mm and 0.50 mm for the manual measurements. The interclass correlation coefficients for intraobserver variability across all four reviewers (i.e., the correlation between the first and second automated measurements for each reviewer) were 0.98-1.0. Similarly, the interclass correlation coefficients describing the interobserver variability between the first automated measurements for reviewers 3 and 4 were 0.98-1.0, compared with 0.94-1.0 for manual measurements.
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The precision of the automated measurement tool was extremely high between both multiple measurements made by a single observer and measurements made by different observers. All four reviewers in our study measured each polyp twice using the automated polyp measurement tool by mouse-clicking at different points on the surface of the polyp. The difference between these two automated measurements was 0.1 mm or less 82-93% of the time, indicating that there is negligible click-point dependency associated with polyp measurement. Furthermore, automated measurements made by two observers were more precise than manual measurements 85% of the time (p < 0.01). Given this high precision, automated polyp measurement may be particularly useful in gauging polyp growth when a polyp is followed with CT colonography.
In previous phantom studies with CT colonographic workstations, investigators examined manual techniques of polyp measurement and found that optimal measurement methods likely depend on the computer workstation used [13, 14]. Young et al. [13] found that pedunculated polyps were measured more accurately than flat polyps with manual techniques, in agreement with our findings with an automated measurement tool.
Automated polyp measurement has been examined in two previous phantom studies [15, 16]. Although the effect of polyp morphologic type or location on haustral folds on accuracy of automated measurement was not examined in either of these studies, Blake et al. [15] predicted our finding that haustral folds may limit the utility of automated measurements in some instances. We found that for some polyps on the tip of a haustral fold, the automated polyp measurement tool circumscribed a portion of the fold tip in addition to the polyp itself, including the fold in the measurement and causing overestimation of polyp size. Like Burling et al. [16], we found that automated measurements were more precise than manual measurements but also found that the accuracy of automated polyp measurements depended on the morphologic type and location of the polyp.
There were several weaknesses to our study. We relied on a rigid colon phantom that presented a limited number of geometric configurations of polyps and folds and did not contain feces or simulate suboptimal inflation. Our polyps were composed of epoxy resins and powders (Solid Water), not soft tissue, which allowed for milling to precise sizes. The fact that the polyps were composed of epoxy whereas the colonic wall was composed of glass may have aided the manual measurements because there is an attenuation difference between these two materials. Furthermore, none of our flat polyps was completely flat. The flat polyps had a height one half the width, whereas flat polyps in clinical practice can have heights of 1-2 mm, regardless of size. It will be difficult to detect the edges of such lesions with automated methods.
Another limitation was that we performed manual measurements with 2D methods. The Siemens software used does not allow 3D measurement of the endoluminal rendering. However, the 2D method used on the workstation compared favorably with 3D measurements obtained with two other commercial workstations [13]. Finally, polyps are deformable, size depending on factors such as pathologic features, geometric configuration, the effect of gravity, colonic inflation, and rotation. The ability to acquire precise polyp measurements between two CT examinations with any manual or automated system depends on the complex interaction of these parameters. It also depends on measurement technique and the limits of reproducibility of CT colonographic images for reliable similar display of polyps at each examination. This comparison can only be performed on patients undergoing CT colonography twice within a short time. We performed a methodologic analysis of accuracy and precision on individual polyps. In the clinical situation, however, data sets are compared side by side.
In conclusion, the accuracy of an automated polyp measurement tool can be affected by polyp size, morphologic type, and location with respect to a haustral fold. When using an automated tool, radiologists should visually inspect automated polyp measurements to ensure they do not include adjacent haustral folds, particularly for small and flat polyps. Manual measurements are more accurate in this setting and may be required when an automated polyp measurement tool yields incorrect polyp boundaries. The precision of polyp measurements obtained with an automated tool is superior to that of manual measurements and may be particularly helpful in assessing interval polyp growth on CT colonography.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  R. M. Summers, J. Liu, J. Yao, L. Brown, J. R. Choi, and P. J. Pickhardt Automated Measurement of Colorectal Polyp Height at CT Colonography: Hyperplastic Polyps Are Flatter Than Adenomatous Polyps Am. J. Roentgenol., November 1, 2009; 193(5): 1305 - 1310. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
