Published online before print November 21, 2007, 10.1148/radiol.2453061930
(Radiology 2008;246:157-167.)
© RSNA, 2007
Linear Polyp Measurement at CT Colonography: 3D Endoluminal Measurement with Optimized Surface-rendering Threshold Value and Automated Measurement1
Seong Ho Park, MD,
Eugene K. Choi, MD,
Seung Soo Lee, MD,
Ji Young Woo, MD,
Soo Young Chung, MD,
Young Jun Kim, MD,
Joon Koo Han, MD, and
Hyun Kwon Ha, MD
1 From the Department of Radiology and Research Institute of Radiology, University of Ulsan College of Medicine, Asan Medical Center, 388-1 Poongnap-Dong, Songpa-Gu, Seoul 138-040, Korea (S.H.P., S.S.L., H.K.H.); Weill Medical College of Cornell University, New York, NY (E.K.C.); Department of Radiology, Hallym University College of Medicine, Kangnam Sacred Heart Hospital, Seoul, Korea (J.Y.W., S.Y.C.); Department of Radiology, Konkuk University School of Medicine, Konkuk University Hospital, Seoul, Korea (Y.J.K.); and Department of Radiology and Institute of Radiation Medicine, Seoul National University College of Medicine, Seoul National University Hospital, Seoul, Korea (J.K.H.). Received November 12, 2006; revision requested January 15, 2007; revision received January 30; accepted March 16; final version accepted May 7.
Address correspondence to S.H.P. (e-mail: seongho{at}amc.seoul.kr).
 |
ABSTRACT
|
|---|
Purpose: To determine the optimal surface-rendering threshold value for three-dimensional (3D) endoluminal computed tomographic (CT) colonographic images for accurate manual polyp measurement, with direct measurement of simulated polyps as the reference standard, and to assess the agreement between manual 3D measurements and automated measurements.
Materials and Methods: Institutional review board approval was not required for the experimental study with pig colons obtained at an abattoir but was obtained for the use of patient data, with waiver of informed consent. Eighty-six simulated polyps (reference size, 3–15 mm) and 14 human polyps (approximate size, 5–20 mm) were included. Automated polyp measurements and manual measurements with endoluminal views that were surface rendered at threshold values of –800, –700, –600, and –500 HU were performed by one observer. Agreement between CT colonographic measurements and reference sizes and between manual and automated measurements were assessed by using the Bland-Altman method.
Results: For simulated polyps, mean measurement difference between the observed size and reference size was 0.86 mm (95% limits of agreement: –0.52 mm, 2.24 mm), 0.55 mm (95% limits of agreement: –0.75 mm, 1.85 mm), 0.20 mm (95% limits of agreement: –1.11 mm, 1.50 mm), and –0.08 mm (95% limits of agreement: –1.43 mm, 1.27 mm) for –800, –700, –600, and –500 HU, respectively. Mean measurement difference was 0.09 mm (95% limits of agreement: –1.49 mm, 1.67 mm) for automated measurement. Manual polyp size at –500 HU (P = .277) and automated polyp size (P = .288) were not significantly different from reference size. For human polyps, 10 polyps, excluding four lesions that were large, lobulated, or located adjacent to an edge of the haustral fold, showed accurate automated demarcation of lesion boundaries. Automated measurements of the 10 polyps showed the closest agreement with manual measurements at –500 HU.
Conclusion: The optimal surface-rendering threshold value for accurate polyp measurement is approximately –500 HU. Automated measurements agree closely with manual measurements at the optimal threshold value for well-circumscribed smooth rounded polyps.
© RSNA, 2007
|
INTRODUCTION
|
|---|
Computed tomographic (CT) colonography is increasingly used as a screening tool for colorectal cancer. Accurate polyp measurement at CT colonography is of considerable clinical importance because polyp size serves as a rough surrogate of the risk of carcinoma (1) and patient treatment is largely based on polyp size. Although the results may be vendor specific, researchers in several studies have investigated the accuracy of CT colonography in polyp measurement. Findings in a recent study suggested that the accuracy of measurement on three-dimensional (3D) endoluminal views is superior compared with measurement by using standard orthogonal two-dimensional (2D) views (ie, transverse, coronal, and sagittal planes) (2). Results of a subsequent in vitro study, however, indicated that 2D measurement off an optimized multiplanar reformatted (MPR) plane (ie, an arbitrary MPR plane that allows viewing of the maximum polyp diameter) was more accurate than 3D endoluminal measurement (3).
Although the current consensus proposal (1) recommends that the single greatest dimension of the polyp should be obtained regardless of the use of 2D or 3D images, 3D endoluminal measurement is likely to have the advantage of being more suited in finding the long axis of the polyp, as one may instinctively visualize the long axis of the polyp by using the endoluminal view. The standard orthogonal 2D views, on the other hand, may fail to include the long axis (2). Finding an optimized MPR plane for each and every polyp is cumbersome and impractical for clinical practice. Three-dimensional endoluminal measurement, therefore, should become a very effective tool for polyp size estimation with further improvement in measurement accuracy.
Of the various factors that affect the accuracy of polyp measurement on endoluminal images, the threshold value for endoluminal rendering is most likely a critical factor. There is a gradual transition of attenuation values over several voxels at the colonic lumen-wall interface from the attenuation of air to that of soft tissue (4). A lower threshold value will shift the reconstructed luminal surface toward the air side, with a resulting expansion of rendered structures. Polyp size is, therefore, expected to increase with the decrease in the surface-rendering threshold value. Although this relationship between object size and threshold value is well-established (4–7), to our knowledge, in no formal study has the optimal threshold value for endoluminal visualization of CT colonographic images for accurate polyp measurement yet been investigated.
Automating the polyp measurement at CT colonography has been explored with the hope of reducing measurement error, observer variability, and human labor (8–12). Results of studies (10,12) have suggested the superior reliability and accuracy of fully automated measurement in in vitro settings, although the results may be vendor specific. A more rigorous evaluation of the automated measurement would require comparison with optimized manual measurement and validation by using in vivo human polyps. To this end, the purpose of our study was to determine the optimal surface-rendering threshold value for 3D endoluminal CT colonographic images for accurate polyp measurement, with direct measurement of simulated polyps as the reference standard, and to assess the agreement between manual 3D measurements and automated measurements.
|
MATERIALS AND METHODS
|
|---|
Institutional review board approval was not required for the experimental study with pig colons obtained at an abattoir but was obtained for the use of patient data, with waiver of patient informed consent.
Polyps and Patients
Figure 1 shows the flow diagram of polyps and patients in the study.
Pig colonic specimens with simulated polyps and reference standard.—Eighteen pig colonic specimens, each approximately 55–60 cm in length, were prepared from fresh pig colons that were commercially available at an abattoir (Fig 2). Five simulated polyps were made in each colonic specimen by two authors (E.K.C. and J.K.H.) who used a previously described method (13) of puckering and securing the colonic mucosa with 5-0 silk sutures. On the basis of a pilot experiment by the authors, the number of simulated polyps permitted the detection of a 0.5-mm difference in polyp size between CT colonographic measurement and the reference measurement, with a type I error of .0102 (ie, Bonferroni adjustment) and a power of approximately 99.9%.
View larger version (106K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2: Pig colonic phantom consisting of pig colonic specimen of slightly sigmoid configuration (arrowheads) with simulated polyps placed in an acrylic container filled with a mixture of 20 L of 100% soybean oil and 20 mL of iodized oil to simulate the attenuation of abdominal fat (ie, approximately –110 HU). A rectal catheter (arrow) used for colonic insufflation during CT colonography was placed through the simulated anus.
|
|
The simulated polyp resembled a rounded sessile polyp with a narrow neck and slight asymmetry, if any (Fig 3). The five polyps were arranged to be linearly located approximately 10 cm apart from each other. The order of polyp size, categorized into 5 mm or smaller, 6–9 mm, and 10 mm or larger, in each colon was randomized. The number of polyps of each size category in each colonic specimen was not set to remain uniform. With the reference standard, polyp size was obtained in millimeters by means of direct measurement of the maximum diameter of the polyp by an author (E.K.C.) after careful examination of the entire lesion morphology. The colonic specimen was assembled in an acrylic container to constitute the entire phantom (Fig 2). Further details of the colonic phantom are described elsewhere (3).
View larger version (87K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3a: Automated and manual measurements of an 8-mm-diameter simulated polyp. (a) Three-dimensional endoluminal view with yellow rectangular parallelepiped and (b) 2D MPR view with the rectangle outline represent the automated demarcation of the polyp boundary. The polyp is closely contained within the rectangular parallelepiped and rectangle, resulting in an accurate measurement of 8 mm. (c) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal view by the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear size of the polyp at –800 HU. The polyp was 9 mm at –800 HU and 8 mm at –500 HU. Figures were captured with the same viewing angle to clearly demonstrate the effect of threshold value on polyp size, but the measurement could be performed after free rotation.
|
|
View larger version (85K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3b: Automated and manual measurements of an 8-mm-diameter simulated polyp. (a) Three-dimensional endoluminal view with yellow rectangular parallelepiped and (b) 2D MPR view with the rectangle outline represent the automated demarcation of the polyp boundary. The polyp is closely contained within the rectangular parallelepiped and rectangle, resulting in an accurate measurement of 8 mm. (c) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal view by the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear size of the polyp at –800 HU. The polyp was 9 mm at –800 HU and 8 mm at –500 HU. Figures were captured with the same viewing angle to clearly demonstrate the effect of threshold value on polyp size, but the measurement could be performed after free rotation.
|
|
View larger version (29K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3c: Automated and manual measurements of an 8-mm-diameter simulated polyp. (a) Three-dimensional endoluminal view with yellow rectangular parallelepiped and (b) 2D MPR view with the rectangle outline represent the automated demarcation of the polyp boundary. The polyp is closely contained within the rectangular parallelepiped and rectangle, resulting in an accurate measurement of 8 mm. (c) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal view by the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear size of the polyp at –800 HU. The polyp was 9 mm at –800 HU and 8 mm at –500 HU. Figures were captured with the same viewing angle to clearly demonstrate the effect of threshold value on polyp size, but the measurement could be performed after free rotation.
|
|
In vivo human polyps.—The in vivo human polyp data set consisted of 14 colorectal polyps (size range, 5–20 mm; measured at optical colonoscopy by comparison with open-biopsy forceps [n = 12] or at surgery as a specimen [n = 2]). Eleven polyps were sessile and three were pedunculated. The polyps included one serrated adenoma, seven tubular adenomas, one tubular adenoma with high-grade dysplasia, and five adenocarcinomas. These polyps were in 10 patients (nine men, one woman; age range, 55–72 years; mean age, 63.8 years) who were chosen from 114 consecutive clinical patients who had undergone CT colonography at Asan Medical Center, Seoul, Korea, during April 2006, with consensus of two authors (S.H.P., with experience of more than 500 cases, and S.S.L., with experience of more than 100 cases). The following inclusion criteria were used: polyp location within a well-distended well-cleansed air-filled portion of the colon, absence of any image artifacts, a clear match at optical colonoscopy, and available pathologic diagnosis.
CT Colonography and Interpretation
CT was performed with a 16–detector row scanner (Somatom Sensation 16; Siemens Medical Solutions, Erlangen, Germany) with the following parameters: beam collimation, 16 x 0.75 mm; section thickness, 1 mm; reconstruction interval, 0.7 mm; beam pitch, 1; gantry rotation time, 0.5 second; field of view, 35 cm for the pig colonic specimens and a size to fit for patients; 120 kV; and 50 mAs. Manual 3D endoluminal measurement of the polyps was performed by using a commercial CT colonographic system (AW4.2_06; GE Healthcare, Waukesha, Wis) with the surface rendering parameters of a smooth rendering mode, 90° of aperture, a wet color mode, and a user-definable threshold value for surface rendering.
Automated measurement was performed by using the automated function of a commercial CT colonographic system (Syngo Colonography PEV; Siemens Medical Solutions). The automated function requires an initial mouse click on a polyp by the reader. The information of polyp edge is then obtained not by using threshold values but by computing the gradient along the polyp border and determining the midpoint of edge transition. The surface area associated with the polyp is determined by computing the patterns of curvature along the surface in proximity of the clicked location and searching for the places of curvature inflection to determine the location of the attachment of the polyp to the colonic wall. The portion underneath the polyp is then interpolated by performing a surface fitting in spherical space (14).
Once the volume associated with the polyp has been determined, the Eigen values and vectors are computed to determine the principal axis of the polyp. Then, the actual extent of the encapsulating rectangular parallelepiped is computed, making sure that the polyp is completely contained. Three-dimensional endoluminal views of the polyp with an overlay of the encapsulating rectangular parallelepiped (Fig 3a) and 2D MPR views of the polyp with an overlay of the rectangle that represents the cut plane of the rectangular parallelepiped (Fig 3b) are presented on the screen to allow reader confirmation. Two CT colonographic systems were used because the former (AW4.2_06; GE Healthcare) did not have the automated function, whereas the latter (Syngo Colonography PEV; Siemens Medical Solutions) did not have the capability to perform 3D measurement.
A radiologist (observer 1, S.H.P.) obtained measurements of the simulated and human polyps by using the automated software and by manually measuring them at four empirically chosen surface-rendering threshold values of –800, –700, –600, and –500 HU. Measurement at each threshold value was performed at a separate session, with a 1-week interval between measurements. The reader was blinded to the threshold value and the size range and specific size of polyps but was informed of the location of each polyp. The reader was free to rotate images and examine the lesion morphology carefully before determining the maximum polyp diameter. Care was particularly taken to avoid measuring beyond the edge of the polyp during the manual 3D measurements. All measurements were rounded to the nearest millimeter.
With the automated measurement, the reader visually evaluated the extent of containment of the polyp within the rectangular parallelepiped and rectangle (ie, software-determined lesion boundary) and recorded any macroscopic error of lesion boundary determination.
To assess interobserver agreement of polyp measurements, another radiologist (observer 2, S.S.L.), who was blinded to the threshold value and the size range and specific size of polyps, measured the polyps manually at the threshold values of –800 and –500 HU (ie, the lowest and the highest threshold values, respectively) and by using the automated software.
Statistical Analysis
After CT scanning, the colon was disassembled and inverted inside out. The simulated polyps were remeasured by an author (E.K.C.). Four simulated polyps that showed size differences between the initial measurement and the remeasurement were excluded. The remaining 86 simulated polyps (25 polyps of 3–5 mm, 38 polyps of 6–9 mm, and 23 polyps of 10–15 mm; mean size, 7.5 mm ± 2.8 [standard deviation]) were included in the analysis.
Interobserver agreement.—Interobserver agreement of polyp measurement at CT colonography was assessed by using the Bland-Altman method (15). The 95% limits of agreement were used to define the range of discrepancies between the two observers on 95% of occasions.
Measurement accuracy and agreement between manual and automated measurements for in vitro simulated polyps.—The Bland-Altman method was used to assess the agreement between polyp measurement at CT colonography and the reference polyp size. The distribution of measurement discrepancies across polyp size was examined with Bland-Altman plots. The Spearman correlation analysis between the reference polyp size and absolute value of measurement discrepancy was performed. The paired t test was used to determine the presence of a significant difference between each set of CT colonographic measurements and the reference size. Change in polyp size categorization (ie, 
5 mm, 6–9 mm, and 
10 mm) as a result of change in the surface-rendering threshold value for manual measurement was assessed. The agreement between manual measurement at each threshold value and automated measurement was assessed by using the Bland-Altman method.
Agreement between manual and automated measurements for in vivo human polyps.—The precise reference sizes of the in vivo human polyps were not obtainable because of the innate inaccuracy of optical colonoscopic measurement and specimen measurement after polyp removal (2,16). It was, therefore, not feasible to directly estimate the error of CT colonographic measurement. Thus, the polyp size measured at CT colonography was compared between different surface-rendering threshold values. Change in polyp size categorization with a change in threshold value was assessed. For those human polyps with boundaries that were accurately determined by using the automated software, the agreement between manual measurement at each threshold value and the automated measurement was assessed by using the Bland-Altman method.
Statistical analysis was performed with software (SPSS 11.5; SPSS, Chicago, Ill). A difference with a P value of less than .0102 (ie, Bonferroni adjustment to account for multiple comparisons) was considered to be statistically significant.
|
RESULTS
|
|---|
Interobserver agreement for the manual and automated measurements at CT colonography is summarized in Table 1.
Measurement accuracy and agreement between manual and automated measurements for in vitro simulated polyps.—There was a general decline in polyp size with manual measurement as the threshold value changed from –800 to –500 HU (Fig 3c), as demonstrated by the gradual shift of the mean measurement difference between the observer and reference polyp size (ie, observer reference, from 0.86 mm at –800 HU to –0.08 mm at –500 HU) (Table 2). The measurement difference was unrelated to polyp size, as demonstrated by no convergence or divergence of data points in Bland-Altman plots (Fig 4, A–D) and no statistically significant correlation between the reference polyp size and absolute value of the measurement difference for –800 HU (r = –.200, P = .064), –700 HU (r = –.122, P = .261), –600 HU (r = .162, P = .135), and –500 HU (r = .233, P = .031). Polyp size measured manually at threshold values of –800 HU (P < .0005), –700 HU (P < .0005), and –600 HU (P = .007) was significantly different from the reference size, whereas polyp size at –500 HU (P = .277) was not significantly different from the reference size (Table 2). Compared with the polyp size categories according to the measurement at –800 HU, eight, 15, and 18 polyps were reclassified into the next smaller size category when they were measured at –700, –600, and –500 HU, respectively (Table 3).
View this table:
[in this window]
[in a new window]
|
Table 2. Agreement between Polyp Measurement at CT Colonography and Reference Polyp Size in in Vitro Simulated Polyps
|
|
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4: Bland-Altman plots show the distribution of polyp measurement difference between the observer and reference polyp sizes (ie, reference size subtracted from observer size, across the reference polyp sizes). Each petal on the data point represents one polyp. Those data points without petals represent a single polyp. The measurement difference is unrelated to polyp size, as demonstrated by no convergence or divergence of data points across the reference polyp size in the plots: manual measurement at A, –800 HU; B, –700 HU; C, –600 HU; and D, –500 HU; and E, automated measurement.
|
|
In terms of automated measurement, macroscopic error in lesion boundary determination did not occur in any simulated polyps (Fig 3a, 3b). The mean measurement difference between the observer and reference polyp size was 0.09 mm, and the 95% limits of agreement were –1.49 to 1.67 mm, respectively (Table 2). The measurement difference was unrelated to polyp size, as demonstrated by no convergence or divergence of data points in Bland-Altman plots (Fig 4, E) and no statistically significant correlation between the reference polyp size and absolute value of the measurement difference (r = .040, P = .716). The automated polyp measurement was not significantly different from the reference size (P = .288) (Table 2). The agreement between automated measurement and manual measurement at each threshold value is summarized in Table 4.
View this table:
[in this window]
[in a new window]
|
Table 4. Agreement between Manual and Automated Polyp Measurement at CT Colonography in in Vitro Simulated Polyps
|
|
Agreement between manual and automated measurements for in vivo human polyps.—There was a general decline in polyp size with the manual measurement as the threshold value changed from –800 HU to –500 HU (Fig 5a). The mean size of the 14 human polyps was 12.0 mm ± 4.8 at –800 HU, 11.7 mm ± 5.1 at –700 HU, 11.5 mm ± 5.0 at –600 HU, and 10.9 mm ± 4.8 at –500 HU. Compared with the polyp size categories according to the measurement at –800 HU, one, two, and three polyps were reclassified into the next smaller size category when they were measured at –700, –600, and –500 HU, respectively (Table 5).
View larger version (36K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5a: Manual and automated measurements of a sessile tubular adenoma in the cecum of a 72-year-old man. (a) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal views with the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear polyp size at –800 HU. The polyp was 6 mm at –800 HU and 5 mm at –500 HU. Measurement could be performed after free rotation, unlike with the presented figures. (b) Three-dimensional endoluminal view shows yellow rectangular parallelepiped and (c) 2D MPR view shows rectangle that represent the polyp boundaries as determined by using the software. The polyp boundaries are extended to a bulbous edge of the haustral fold (arrowheads), resulting in an erroneous size overestimation of 9 mm.
|
|
View larger version (102K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5b: Manual and automated measurements of a sessile tubular adenoma in the cecum of a 72-year-old man. (a) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal views with the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear polyp size at –800 HU. The polyp was 6 mm at –800 HU and 5 mm at –500 HU. Measurement could be performed after free rotation, unlike with the presented figures. (b) Three-dimensional endoluminal view shows yellow rectangular parallelepiped and (c) 2D MPR view shows rectangle that represent the polyp boundaries as determined by using the software. The polyp boundaries are extended to a bulbous edge of the haustral fold (arrowheads), resulting in an erroneous size overestimation of 9 mm.
|
|
View larger version (71K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5c: Manual and automated measurements of a sessile tubular adenoma in the cecum of a 72-year-old man. (a) The gradual decrease in polyp size with the increase in threshold value from –800 HU to –500 HU is clearly demonstrated on the 3D endoluminal views with the comparison of the polyp size with the length of the line overlaid on the polyp that represents the linear polyp size at –800 HU. The polyp was 6 mm at –800 HU and 5 mm at –500 HU. Measurement could be performed after free rotation, unlike with the presented figures. (b) Three-dimensional endoluminal view shows yellow rectangular parallelepiped and (c) 2D MPR view shows rectangle that represent the polyp boundaries as determined by using the software. The polyp boundaries are extended to a bulbous edge of the haustral fold (arrowheads), resulting in an erroneous size overestimation of 9 mm.
|
|
In terms of automated measurement, the lesion boundary was apparently inaccurately determined in four polyps. In one sessile polyp, size overestimation occurred because of the misrecognition of a bulbous edge of the haustral fold adjacent to the polyp as a part of the lesion (Fig 5b, 5c), and this size overestimation resulted in an automated size of 9 mm in comparison with manual sizes of 6 mm at –800 HU and 5 mm at –500 HU. An underestimation occurred in three sessile polyps with lobulated and irregular contours, where recognition of only a portion of the lesions was made (Fig 6), and this underestimation resulted in automated sizes of 5 mm, 14 mm, and 16 mm, respectively, for lesions for which the manual sizes were 14 mm, 17 mm, and 19 mm at –800 HU, respectively, and 13 mm, 16 mm, and 18 mm at –500 HU, respectively. Automated demarcation of lesion boundary was accurate in the remaining 10 polyps (seven sessile and three pedunculated) that were well circumscribed, smooth, and rounded. In those 10 polyps, the automated measurement demonstrated the closest agreement with the manual measurement at –500 HU (Table 6).
View larger version (158K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6a: Automated measurement of a sessile adenocarcinoma in the sigmoid colon of a 55-year-old man. (a) Three-dimensional endoluminal CT colonographic view shows sessile morphology of the lesion (arrow), which is located at an obtuse angle to the adjacent mucosa. The central portion of the lesion protrudes farther into the lumen than the surrounding periphery, which produces a polyp-in-polyp appearance. (b) Three-dimensional endoluminal view shows only the central portion of the lesion that is included within the yellow rectangular parallelepiped and (c) 2D MPR view shows the rectangle. Measurement resulted in an erroneous size underestimation of 5 mm. The lesion was 14 mm at –800 HU and 13 mm at –500 HU.
|
|
View larger version (100K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6b: Automated measurement of a sessile adenocarcinoma in the sigmoid colon of a 55-year-old man. (a) Three-dimensional endoluminal CT colonographic view shows sessile morphology of the lesion (arrow), which is located at an obtuse angle to the adjacent mucosa. The central portion of the lesion protrudes farther into the lumen than the surrounding periphery, which produces a polyp-in-polyp appearance. (b) Three-dimensional endoluminal view shows only the central portion of the lesion that is included within the yellow rectangular parallelepiped and (c) 2D MPR view shows the rectangle. Measurement resulted in an erroneous size underestimation of 5 mm. The lesion was 14 mm at –800 HU and 13 mm at –500 HU.
|
|
View larger version (77K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6c: Automated measurement of a sessile adenocarcinoma in the sigmoid colon of a 55-year-old man. (a) Three-dimensional endoluminal CT colonographic view shows sessile morphology of the lesion (arrow), which is located at an obtuse angle to the adjacent mucosa. The central portion of the lesion protrudes farther into the lumen than the surrounding periphery, which produces a polyp-in-polyp appearance. (b) Three-dimensional endoluminal view shows only the central portion of the lesion that is included within the yellow rectangular parallelepiped and (c) 2D MPR view shows the rectangle. Measurement resulted in an erroneous size underestimation of 5 mm. The lesion was 14 mm at –800 HU and 13 mm at –500 HU.
|
|
|
DISCUSSION
|
|---|
Our results showed that the polyp size was substantially affected by the surface-rendering threshold value for endoluminal visualization at CT colonography. There was a general decline in polyp size with an increase in threshold value from –800 to –500 HU, and this pattern was in accordance with what was expected on the basis of the principles of surface rendering. In CT colonographic screening protocols where polyp size dictates clinical treatment, this overlooked source of measurement error caused by surface-rendering threshold values for 3D visualization can cause problems. The clinical consequence of this error is overt, especially in patients who have polyps with sizes that are on the borderline between size categories (ie, 5 mm, 6–9 mm, and 10 mm); in these patients, a slight size disparity may lead to changes in polyp size categorization, as indicated with our data.
According to the results from simulated polyps, polyp measurement obtained with –500 HU was shown to be the most accurate of the four surface-rendering threshold values for the specific CT colonographic system tested. The optimal threshold value for 3D endoluminal surface rendering for accurate polyp measurement, therefore, is suggested to be approximately –500 HU. This conclusion cannot be definitely drawn, however, from data derived from in vivo human polyp measurements because of the technical infeasibility (2,16) of obtaining the precise reference size and the small number of lesions included. However, the fact that automated measurement of the 10 human polyps (with boundaries that were accurately determined by using the automated software) was very close to the manual measurement at –500 HU, as it was with the simulated polyps, may provide indirect evidence that the conclusion should also hold for human polyps. The 95% limits of agreement between the automated and manual measurements were wider for the human polyps than they were for the simulated polyps, and this difference suggests a larger variability in measurement in the in vivo setting compared with the in vitro experimental setting. An optimal surface-rendering threshold value, therefore, may require further confirmation in human polyps.
In contrast to the fairly uniform method of 2D display at CT colonography across vendors, various CT colonographic systems may use different techniques for endoluminal visualization (17). Some technical details are proprietary. Our study is limited in that the results may be vendor specific and may not be directly applicable to other systems, especially to those systems with which the volume-rendering techniques are used. Volume-rendering display involves a more complex set of choices than simple threshold. Nevertheless, with the volume-rendering technique, images are reconstructed on the basis of a predetermined range of attenuation values that represents the colonic mucosal surface and thus the same principle applies; that is, the selection of the range of attenuation values may substantially affect polyp size measurement. Although confirmation with further studies is needed, we conjecture that attenuation values around –500 HU may be an optimal attenuation range for the volume-rendered reformatted image of the colon for accurate polyp measurement.
Various CT colonographic systems with their own 3D rendering techniques may produce variable degrees of error with 3D measurements. With the proliferation of CT colonography as a noninvasive screening tool for colorectal cancer, uniform interpretation and patient treatment based on polyp measurement by using various CT colonographyic systems is important. To achieve this uniformity, the implementation of a standardized calibration tool, such as a standard polyp phantom, may be required to help optimize the rendering parameters of each CT colonographic system to ensure accurate and reproducible 3D polyp measurements across various CT colonographic systems. Surveillance for polyp size progression by using CT colonography without polypectomy is not an established practice yet (18,19). However, if we are to use CT colonography and its 3D polyp measurement for noninvasive follow-up of the polyp size, optimization of the 3D visualization of CT colonographic systems to achieve consistent and accurate polyp size comparison between examinations is important for proper patient treatment.
Of late, automation of polyp measurements has received attention as a potential method of reducing measurement error, observer variability, and human labor (8–12). Our results corroborate the feasibility of the automated measurement technique by demonstrating that the measurements with the CT colonographic system evaluated in our study showed close agreement with the manual 3D measurements at the optimal surface-rendering threshold value in simulated polyps and in well-circumscribed smooth rounded human polyps. The automated measurement also showed complete reproducibility in our study with both in vitro and in vivo conditions. The limits of agreement with the reference polyp size, however, were slightly wider for the automated measurement compared with the manual measurement, which suggests a lower precision of the automated measurement compared with the manual measurement. Misidentification of the polyp boundary, leading to erroneous measurements of some human polyps, also highlights a major limitation of the automated technique. Polyps that were large, lobulated, irregularly shaped, or located adjacent to a bulbous edge of the haustral fold were especially likely to have boundaries that were misidentified. The existence of such potential limitations precludes the possibility that automated measurement can entirely supplant the manual approach. Automated measurement, however, will be very useful when it is accompanied by visual confirmation by a human reader and manual remeasurement in cases in which error is suspected.
Our study had limitations. First, polyp measurement accuracy at CT colonography may be affected by factors other than threshold values, such as lesion location in relation to haustral folds and colonic curvature, degree of colonic distention (20), lesion morphology, noise (ie, radiation dose), and reader experience (21,22). All the polyps in our study, however, were located in a well-distended, well-cleansed, air-filled portion of the colon and all simulated polyps were rounded sessile lesions located on the haustral surface of near straight segments of the colon. Moreover, data were obtained by using a single radiation dose and a single experienced reader. Nevertheless, the study was specifically designed to determine the effects of surface-rendering threshold values on polyp measurements, and thus other possible confounding factors that could potentially affect measurement were intentionally controlled.
Second, given the fact that the reference measurement and each manual CT colonographic measurement at each threshold value were performed independently, the long axis of the polyp was determined separately for each measurement. Any slight variation in the identification of the long axis of the polyp at each measurement would have confounded the effect of varying threshold values on polyp size and would have affected the agreement between the polyp size at CT colonography and the reference size. The results may have been more precise if the vector of the long axis of the polyp had been recorded at each measurement. We believe, however, that this factor had a minimal effect because the entire lesion morphology was carefully reviewed before any measurements were obtained.
Third, the results may have been more scientifically precise and informative if the radiologist had obtained several polyp measurements by using CT colonography and then subsequently had determined the mean of the values to calculate a final measurement with an associated standard deviation. This systematic method, however, does not represent the way a polyp is measured in clinical practice.
Fourth, we did not include patients who underwent intravenous contrast material–enhanced CT colonography. Although the use of intravenous contrast material may change the threshold value analysis results, its use is not a standard procedure for screening CT colonography.
In conclusion, the optimal threshold value for endoluminal surface rendering at CT colonography for accurate polyp measurement is approximately –500 HU. Automated polyp measurement is reproducible and agrees closely with manual endoluminal measurement at the optimal threshold value for well-circumscribed smooth rounded polyps. These findings may require further confirmation in a study in a large number of human polyps.
|
ADVANCES IN KNOWLEDGE
|
|---|
- Polyp size measured on three-dimensional endoluminal views with CT colonography is substantially affected by the threshold value for surface rendering.
- The optimal surface-rendering threshold value for accurate polyp measurement by using endoluminal views is approximately –500 HU, although it may be vendor specific.
- Automated measurement with the CT colonographic system evaluated is completely reproducible and agrees closely with the manual measurement at the optimal surface-rendering threshold value for well-circumscribed smooth rounded polyps; however, such measurement is prone to macroscopic error especially in polyps that are large, lobulated, irregularly shaped, or located adjacent to a bulbous edge of the haustral fold.
|
IMPLICATION FOR PATIENT CARE
|
|---|
- The results of our study may help to ensure accurate and consistent measurement of colorectal polyps at CT colonography and thereby promote appropriate treatment of patients with colorectal polyps.
|
ACKNOWLEDGMENTS
|
|---|
The authors thank Luca Bogoni, PhD, Siemens Medical Solutions USA, Malvern, Pa, for providing the information on the algorithm of automated polyp measurement with Syngo Colonography PEV.
|
FOOTNOTES
|
|---|
Abbreviations: MPR = multiplanar reformatted • 3D = three-dimensional • 2D = two-dimensional
Guarantor of integrity of entire study, S.H.P.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, S.H.P., E.K.C.; clinical studies, S.H.P., S.S.L.; experimental studies, S.H.P., E.K.C., S.S.L., J.K.H.; statistical analysis, S.H.P.; and manuscript editing, S.H.P., E.K.C., J.Y.W., S.Y.C., Y.J.K., H.K.H.
Authors stated no financial relationship to disclose.
|
References
|
|---|
- Zalis ME, Barish MA, Choi JR, et al. CT colonography reporting and data system: a consensus proposal. Radiology 2005;236:3–9. [Free Full Text]
- Pickhardt PJ, Lee AD, McFarland EG, Taylor AJ. Linear polyp measurement at CT colonography: in vitro and in vivo comparison of two-dimensional and three-dimensional displays. Radiology 2005;236:872–878. [Abstract/Free Full Text]
- Park SH, Choi EK, Lee SS, et al. Polyp measurement reliability, accuracy, and discrepancy: optical colonoscopy versus CT colonography with pig colonic specimens. Radiology 2007;244:157–164. [Abstract/Free Full Text]
- Hopper KD, Iyriboz AT, Wise SW, Neuman JD, Mauger DT, Kasales CJ. Mucosal detail at CT virtual reality: surface versus volume rendering. Radiology 2000;214:517–522. [Abstract/Free Full Text]
- Lu B, Dai RP, Jiang SL, et al. Effects of window and threshold levels on the accuracy of three-dimensional rendering techniques in coronary artery electron-beam CT angiography. Acad Radiol 2001;8:754–761. [CrossRef][Medline]
- Halpern EJ, Wechsler RJ, DiCampli D. Threshold selection for CT angiography shaded surface display of the renal arteries. J Digit Imaging 1995;8:142–147. [Medline]
- Addis KA, Hopper KD, Iyriboz TA, Kasales CJ, Liu Y, Wise SW. Optimization of shaded surface display for CT angiography. Acad Radiol 2001;8:976–981. [CrossRef][Medline]
- Bielen D, De Wever L, Drisis S, Guendel L, Kiss G. Accuracy of automated size measurements of CT colonography in phantom and patient data [abstr]. In: Radiological Society of North America Scientific Assembly and Annual Meeting Program. Oak Brook, Ill: Radiological Society of North America, 2005; 299.
- Booya F, Fletcher JG, Melton Z, et al. Automated polyp measurement at CT colonography: preliminary observations in a phantom colon model [abstr]. In: Radiological Society of North America Scientific Assembly and Annual Meeting Program. Oak Brook, Ill: Radiological Society of North America, 2005; 299–300.
- Burling D, Halligan S, Roddie ME, et al. Computed tomography colonography: automated diameter and volume measurement of colonic polyps compared with a manual technique—in vitro study. J Comput Assist Tomogr 2005;29:387–393. [CrossRef][Medline]
- Taylor SA, Halligan S, Roddie M, et al. CT colonography: automated diameter measurement of colonic polyps compared with a manual technique—in vitro study [abstr]. In: Radiological Society of North America Scientific Assembly and Annual Meeting Program. Oak Brook, Ill: Radiological Society of North America, 2005; 299.
- Taylor SA, Slater A, Halligan S, Honeyfield L, Roddie M, Dehmeshki J. CT colonography: automated measurement of colonic polyps compared with a manual technique—in vitro study using a human colectomy specimen [abstr]. In: Radiological Society of North America Scientific Assembly and Annual Meeting Program. Oak Brook, Ill: Radiological Society of North America, 2005; 299.
- Dachman AH, Lieberman J, Osnis RB, et al. Small simulated polyps in pig colon: sensitivity of CT virtual colography. Radiology 1997;203:427–430. [Abstract/Free Full Text]
- Jerebko A, Lakare S, Cathier P, Periaswamy S, Bogoni L. Symmetric curvature patterns for colonic polyp detection. Med Image Comput Comput Assist Interv Int Conf Med Image Comput Comput Assist Interv 2006;9(pt 2):169–176. [Medline]
- Bland JM, Altman DG. Statistical method for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–310. [CrossRef][Medline]
- Gopalswamy N, Shenoy VN, Choudhry U, et al. Is in vivo measurement of size of polyps during colonoscopy accurate? Gastrointest Endosc 1997;46:497–502.
- Pickhardt PJ. Three-dimensional endoluminal CT colonography (virtual colonoscopy): comparison of three commercially available systems. AJR Am J Roentgenol 2003;181:1599–1606. [Abstract/Free Full Text]
- Ransohoff DF. CON: immediate colonoscopy is not necessary in patients who have polyps smaller than 1 cm on computed tomographic colonography. Am J Gastroenterol 2005;100:1905–1907. [CrossRef][Medline]
- Rex DK. PRO: patients with polyps smaller than 1 cm on computed tomographic colonography should be offered colonoscopy and polypectomy. Am J Gastroenterol 2005;100:1903–1905. [Medline]
- Taylor S, Slater A, Honeyfield L, Burling D, Halligan S. CT colonography: effect of colonic distension on polyp measurement accuracy and agreement—in vitro study. Acad Radiol 2006;13:850–859. [CrossRef][Medline]
- Burling D, Halligan S, Taylor S, et al. Polyp measurement using CT colonography: agreement with colonoscopy and effect of viewing conditions on interobserver and intraobserver agreement. AJR Am J Roentgenol 2006;186:1597–1604. [Abstract/Free Full Text]
- Burling D, Halligan S, Altman DG, et al. Polyp measurement and size categorisation by CT colonography: effect of observer experience in a multi-centre setting. Eur Radiol 2006;16:1737–1744. [CrossRef][Medline]
This article has been cited by other articles:
|
|
|
|
 
S. H. Park, S. S. Lee, J. K. Kim, M.-J. Kim, H. J. Kim, S. Y. Kim, M.-Y. Kim, A. Y. Kim, and H. K. Ha
Volume Rendering with Color Coding of Tagged Stool during Endoluminal Fly-through CT Colonography: Effect on Reading Efficiency
Radiology,
September 1, 2008;
248(3):
1018 - 1027.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
