Radiology
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online before print September 27, 2006, 10.1148/radiol.2412051223
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2412051223v1
241/2/426    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by O'Connor, S. D.
Right arrow Articles by Choi, J. R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by O'Connor, S. D.
Right arrow Articles by Choi, J. R.
(Radiology 2006;241:426-432.)
© RSNA, 2006

Gastrointestinal Imaging

CT Colonography with Computer-aided Polyp Detection: Volume and Attenuation Thresholds to Reduce False-Positive Findings Owing to the Ileocecal Valve1

Stacy D. O'Connor, SB, Ronald M. Summers, MD, PhD, Jianhua Yao, PhD, Perry J. Pickhardt, MD and J. Richard Choi, ScD, MD

1 From the Department of Radiology, National Institutes of Health, 10 Center Dr, Bldg 10, Rm 1C351, MSC 1182, Bethesda, MD 20892-1182 (S.D.O., R.M.S., J.Y.); National Naval Medical Center, Bethesda, Md (P.J.P.); Department of Radiology, University of Wisconsin Medical School, Madison, Wis (P.J.P.); Uniformed Services University of the Health Sciences, Bethesda, Md (P.J.P., J.R.C.); and Walter Reed Army Medical Center, Washington, DC (P.J.P., J.R.C.). Received July 22, 2005; revision requested September 23; revision received November 16; accepted December 20; final version accepted February 17, 2006. Supported by the Clinical Research Training Program, a public-private partnership supported jointly by the National Institutes of Health (NIH), and a grant to the Foundation for the NIH from Pfizer Pharmaceuticals Group; and in part by the Intramural Research Program of the NIH, Warren G. Magnuson Clinical Center. Address correspondence to R.M.S. (e-mail: rms{at}nih.gov).


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Purpose: To retrospectively identify volume and average attenuation thresholds for differentiating between ileocecal valve (ICV) and polyp at computed tomographic (CT) colonography with computer-aided detection (CAD).

Materials and Methods: Informed consent (with consent for future retrospective research) and institutional review board (IRB) approval were obtained for the original prospective study. This retrospective study had IRB approval, as well, and was HIPAA-compliant. A total of 496 patients were selected from a larger screening population. CT colonographic images from 394 patients (227 men, 167 women; mean age, 58.0 years; range, 40–79 years) were used as a training set, and images from 102 patients (76 men, 26 women; mean age, 59.8 years; range, 46–79 years) were used as a test set. A series of 2742 volume and attenuation thresholds, for which segmented findings both larger in volume and lower in average attenuation were labeled as ICVs and remaining findings were labeled polyps, were applied to the training set to determine settings with 100% sensitivity for polyp detection and the highest specificity for ICV detection. The optimal settings were then applied to the test set. Significance was assessed with the Fisher exact test, and 95% confidence intervals (CIs) were computed for sensitivity and specificity.

Results: A total of 386 ICVs and 67 adenomatous polyps from the training set and 102 ICVs and 138 adenomatous polyps from the test set could be segmented with a three-dimensional segmentation algorithm. When supine and prone images were counted individually, 746 nonunique ICVs from the training set and 191 from the test set were segmentable. In the training set, a volume of 600 mm3 and an attenuation of 36 HU provided 100% sensitivity (67 polyps; 95% CI: 93%, 100%) and the optimal 83% specificity (618 of 746 ICVs; 95% CI: 80%, 85%). When applied to the test set, this combination provided 97% sensitivity (134 of 138 polyps; 95% CI: 92%, 99%) and 84% specificity (160 of 191 ICVs; 95% CI: 78%, 89%). Differences in sensitivity and specificity in the detection of polyps between the sets were not significant.

Conclusion: Volume and average CT attenuation thresholds can help differentiate most ICVs from true polyps.

© RSNA, 2006


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Computer-aided detection (CAD) is feasible and may increase the sensitivity of polyp detection at computed tomographic (CT) colonography (1–5). CAD systems scan the colon for likely polyps and display the results for consideration by a radiologist. Although this helps find true polyps, the burden of false-positive detections must be considered in terms of increased time and cost of CT colonographic examinations, as well as needless follow-up optical colonoscopy examinations. A systematic approach to reduce false-positive detections with known causes would increase the efficiency of CT colonography with CAD.

The ileocecal valve (ICV) can have a polypoid shape and is a common cause of false-positive findings during CT colonography (6–12). The ICV can be at least partially seen at 87%–98% of CT colonographic examinations and 86% of barium enema examinations (6,13,14). The average normal ICV is large, with a height of 1.7 cm and a width of 2.8 cm (14). ICVs often have a lower internal attenuation than do polyps because they may have lipomatous areas or overall lipohyperplasia (15–18).

Previous investigators (6,8,9,11,14,16) have recognized the importance of eliminating false-positive CT colonographic–CAD detections due to ICVs and have offered manual and automatic ways to recognize them. Johnson and Dachman (7) argued that the ICV should be identified and inspected at all CT colonographic examinations. The purpose of our study was to retrospectively identify volume and average attenuation thresholds for differentiating between the ICV and a potential polyp at CT colonography with CAD.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
The authors who were not on the Viatronix (Stony Brook, NY) medical advisory board had control of the data and the information submitted for publication. R.M.S., P.J.P., and J.Y. have patents pending or awarded in the subject area of this article. R.M.S. and J.Y. received a royalty for a patent and software license (since expired) provided by the National Institutes of Health to Viatronix. Viatronix supplied free research software to R.M.S., P.J.P. is on the medical advisory boards of Viatronix and Medicsight (London, England), and J.R.C. is on the medical advisory boards of Viatronix and Quantitative Imaging (San Bruno, Calif).

Study Group
The study cohort consisted of 496 patients selected from a larger screening population cohort recruited at three institutions (19). Patients were grouped into training and test sets (20). The training set was a random one-third of patients from the larger cohort and consisted of 394 patients (227 men and 167 women; mean age, 58.0 years; age range, 40–79 years) with 68 adenomatous polyps diagnosed at optical colonoscopy and visible at CT colonography (0.5–3.0 cm in diameter; mean ± standard deviation, 0.9 cm ± 0.5; zero to three polyps per patient). The test set consisted of 102 patients (76 men and 26 women; mean age, 59.8 years; age range, 46–79 years) with at least one adenomatous polyp per patient, measuring at least 5 mm, diagnosed at optical colonoscopy and visible at CT colonography in the remaining two-thirds of the original cohort. These patients had 140 adenomatous polyps (0.3–4.2 cm; mean, 0.8 cm ± 0.5; one to five polyps per patient). Because it is more important to keep polyp detections than to exclude ICV detections, the test set had more polyps (all patients in the test set had polyps) and fewer ICVs to better assess the algorithm's performance in the detection of polyps. Written informed consent (including consent for future retrospective research) was obtained from all patients, and the study protocol for the original prospective study was approved by the institutional review board at each center. Our retrospective study had institutional review board approval as well. Both studies were compliant with the Health Insurance Portability and Accountability Act.

CT Examination
Patient preparation.—Patients underwent standard 24-hour colonic preparation with oral administration of 90 mL of sodium phosphate (Fleet 1 preparation; Fleet Pharmaceuticals, Lynchburg, Va) and 10 mg of bisacodyl. As part of their clear liquid diet, patients also consumed 500 mL of barium sulfate (Scan C, Lafayette Pharmaceuticals, Lafayette, Ind; 2.1% by weight) for solid stool tagging and a 120-mL solution of diatrizoate meglumine and diatrizoate sodium (Gastrografin; Bracco Diagnostics, Princeton, NJ) to opacify luminal fluid, in divided doses. The dosing schedule was barium, barium and diatrizoate, and diatrizoate, with approximately 12 hours between each administration of contrast material.

CT.—Colonic distention was achieved by patient-controlled rectal insufflation of room air immediately before scanning. CT was performed while the patient held his or her breath in the supine and prone positions; a four- or eight-section CT scanner was used (LightSpeed or LightSpeed Ultra; GE Healthcare, Waukesha, Wis). The CT technique involved the use of 1.25–2.5-mm collimation, a table speed of 15 mm per second, a reconstruction interval of 1 mm, 100 mAs, and 120 kVp.

ICV and Polyp Identification
The location of each ICV was determined by two readers in consensus (R.M.S. and S.D.O. [a trainee]) and recorded in a data file. With use of software developed in our laboratory, the ICV was identified on transverse CT scans by clicking on a voxel inside (approximately at the center of) it. It was not important that the voxel selected was in the precise center of the ICV, only that the voxel was within the ICV. On most studies, the ICV could be identified by locating the terminal ileum and tracing it to its communication with the cecum. Each polyp was similarly identified by two readers in consensus (R.M.S. and another trainee) by using data from optical colonoscopy to determine the approximate location of each lesion. The two trainees were taught to recognize ICVs and polyps. R.M.S., who has 6 years of experience with CT colonography and CAD, reviewed their findings and a consensus opinion was formed. The same localization procedure was employed by all readers for the training and test sets.

Colon software was provided at no charge by the manufacturer (V3D; Viatronix) and used to identify the section, column, and row coordinates of polyps seen on CT colonographic images. The original CT colonographic data were interpreted with a fully-paid version of the Viatronix software as part of the segmental unblinding process to ascertain ground truth (19).

Data Acquisition
The data file of seed voxels, which consisted of one interior point per ICV or polyp per CT colonographic examination, was downloaded to a prototype polyp segmentation algorithm (21,22). This segmentation algorithm identifies the edges of polyps, masses, and ICVs by using an iterative process involving a knowledge-based deformable surface and fuzzy clustering. The segmentation algorithm reports the total volume (in cubic millimeters) and mean interior attenuation (in Hounsfield units) of the voxels within—but not at the boundary of—the segmented region. CT colonographic images were interpolated and resampled to 0.625 x 0.625 x 1 mm per voxel in polyp and ICV regions of interest. The segmentation algorithm was applied to 128 x 128 x 64-voxel subvolumes centered on the seed voxels.

The segmentation algorithm was applied to the training and test sets, producing lists of ICVs and polyps ("findings") that it could analyze. The findings that it could not segment were eliminated. ICVs that cannot be segmented will not be detected as false-positive polyps by our CAD system (20). Several criteria had to be met before findings were deemed segmentable: First, the boundary between the finding and the lumen had to be convex—that is, at least one-fourth of the boundary pixels had to have a curvature smaller than –0.5 cm–1. Second, the finding had to be larger than 8 voxels. Third, the finding could not be flat (width-to-height ratio of <3.5 for nonflat regions). Voxels with CT attenuations greater than 276 HU were considered to be opacified fluid and, as such, were not included in the segmented ICV or polyp. A threshold value of 276 HU was chosen on the basis of our prior experience in the segmentation of the opacified colon (23).

Establishing and Testing Thresholds
At clinical imaging, ICVs have been found to be large and have low attenuation (14,16). On the basis of this observation, we used the training set to establish a threshold for the mean attenuation and a threshold for the total volume of a segmented object: The segmented object had to have both a lower attenuation and larger volume than the threshold for each of these objects to be considered an ICV. To determine sensitivity and specificity, polyps were classified as positive findings and ICVs as negative findings. Therefore, polyps defined as polyps according to the threshold values were true-positive findings, polyps interpreted as ICVs according to the thresholds were false-negative findings, ICVs defined as polyps were false-positive findings, and ICVs defined as ICVs were true-negative findings. Because each ICV detection that was eliminated would be beneficial, we considered detection on the supine and prone images individually. Hence, specificity was calculated by dividing the number of correctly labeled ICV detections by the total number of ICV detections. Consequently, most ICVs are counted twice for the specificity calculation. As long as one detection (on a supine or prone image) of a polyp remains, it may be found, so we counted each polyp only once for sensitivity calculations. We refer to polyp detection on supine, prone, or both supine and prone images as detection of one "unique" polyp—that is, if a polyp was ruled a true-positive finding on at least one scan (supine or prone), it was considered a true-positive finding. Therefore, sensitivity was the number of unique polyps correctly labeled on at least one scan divided by the total number of unique polyps.

With these criteria in place, we tested 2742 attenuation-volume threshold combinations in the training set to find the combination that would yield 100% sensitivity for polyps with the highest specificity. CT attenuation and volume threshold combinations were established by inspecting a graph of CT attenuation versus volume of ICVs and polyps from the training set. A rectangular region was drawn around the bulk of the values (S.D.O.) and divided into 25 mm3 and 10 HU intervals. We then calculated the sensitivity and specificity for the selected threshold in the testing set.

Statistical Analysis
A P value of less than .05 was considered indicative of a statistically significant difference. The Fisher exact test was used to compare the sensitivities of unmatched data (24). Confidence intervals (CIs) for sensitivity and specificity were computed with an online calculator (25).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 
Training Set
Of the 788 ICVs in the training set (from prone and supine scans of 394 patients), 782 (99%) were identified visually by the readers. Seven hundred forty-six of these 782 ICVs (95%) were segmentable. There were 386 unique segmentable ICVs (Fig 1). Of the 68 adenomatous polyps diagnosed at optical colonoscopy and visible at CT colonography, 67 (99%) were segmentable on at least one scan and 55 were segmentable on two scans, for a total of 122 segmented polyps. The mean volume and mean average attenuation of the segmented ICVs were 1489 mm3 ± 1006 (95% CI: 1417, 1561) and –68 HU ± 60 (95% CI: –72, –63), respectively. The mean volume and mean average attenuation of the segmented polyps were 281 mm3 ± 615 (95% CI: 172, 391) and –52 HU ± 109 (95% CI: –72, –33), respectively (Fig 2).


Figure 1
View larger version (105K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 1: Transverse CT scan obtained with the patient prone shows an ICV (arrow). The image has been rotated 180° to bring the patient into a standard anatomic position. The black dot near the center of the ICV is the seed voxel. The circle is the boundary of the three-dimensional segmentation. White voxels (>276 HU) are removed before the volume and mean attenuation are calculated. The volume and average attenuation for this ICV were 150 mm3 and 29 HU, respectively. C = cecum, TI = terminal ileum.

 

Figure 2
View larger version (20K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2a: Plots of volume versus mean CT attenuation for ICVs and true polyps (TPs) found in the training and test sets. (a) Data from the training set. There were 386 segmentable ICVs (746 segmentable findings) and 67 segmentable true polyps (122 segmentable findings). The dashed line represents the area that contains the 2742 volume and attenuation thresholds evaluated for sensitivity and specificity. The gray circle represents the threshold that resulted in 100% sensitivity with the highest specificity (83%). (b) Data from the test set. There were 102 segmentable ICVs (203 segmentable findings) and 138 segmentable true polyps (253 segmentable findings). The gray circle represents the threshold established from the training set.

 

Figure 2
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
  		Figure 2b: Plots of volume versus mean CT attenuation for ICVs and true polyps (TPs) found in the training and test sets. (a) Data from the training set. There were 386 segmentable ICVs (746 segmentable findings) and 67 segmentable true polyps (122 segmentable findings). The dashed line represents the area that contains the 2742 volume and attenuation thresholds evaluated for sensitivity and specificity. The gray circle represents the threshold that resulted in 100% sensitivity with the highest specificity (83%). (b) Data from the test set. There were 102 segmentable ICVs (203 segmentable findings) and 138 segmentable true polyps (253 segmentable findings). The gray circle represents the threshold established from the training set.

 
Test Set
Two hundred three of the 204 ICVs (99.5%) in the test set were identified visually by the readers; 191 (94%) of these were segmentable. There were 102 unique segmentable ICVs. One hundred thirty-eight of 140 polyps (99%) diagnosed at optical colonoscopy and visible at CT colonography were segmentable on at least one scan and 115 were segmentable on two scans, for a total of 253 segmented polyp detections. The mean volume and mean average attenuation of the segmented ICVs were 1565 mm3 ± 923 (95% CI: 1434, 1696) and –68 HU ± 55 (95% CI: –76, –60), respectively. The mean volume and mean average attenuation of the segmented polyps were 271 mm3 ± 706 (95% CI: 184, 358) and –68 HU ± 107 (95% CI: –81, –55), respectively. The differences in volume and mean average attenuation measurements of the polyps or ICVs between the test and training sets were not significant (P > .1).

Establishing and Testing Thresholds
When the 2742 volume-attenuation threshold combinations were tested with the training set, a combination of 600 mm3 and 36 HU achieved the maximum specificity (618 of 746 ICVs [83%]; 95% CI: 80%, 85%) while maintaining 100% sensitivity (67 of 67 polyps; 95% CI: 93%, 100%). All other threshold combinations had lower specificity with 100% sensitivity or less than 100% sensitivity. When applied to the test set, this combination produced 97% sensitivity (134 of 138 polyps; 95% CI: 92%, 99%) and 84% specificity (160 of 191 polyps; 95% CI: 78%, 89%). The four polyps mislabeled in the test set (false-negative findings) were a carcinoma and three adenomas (measuring 1.0, 1.4, and 2.5 cm). The carcinoma was 4.2 cm and had streak artifact, which could have accounted for its low attenuation. The smallest adenoma was actually located on the ICV and, therefore, both were included in the segmentation. The segmentations of two larger adenomas included voxels in air or fatty tissue near the colonic wall. The differences in the sensitivity and specificity for polyps between the training and test sets were not significant (P > .05).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
ADVANCES IN KNOWLEDGE
 References
 
CAD is feasible for CT colonography and may increase the sensitivity in the detection of polyps (1–4,25). CAD has been used with mammography, where it increases sensitivity and enables radiologists to correctly recommend biopsy procedures in patients with masses on serial mammograms while reducing unnecessary biopsies (26–29). Similarly, CAD has improved radiologists' performance in the detection of lung nodules at chest radiography and CT (30,31).

The ICV is a known cause of false-positive findings at CT colonography with or without CAD (6–8,10–13,32). The ICV is responsible for 14%–20% of false-positive findings at CT colonography with CAD (2,4). The ICV is commonly identifiable and was seen on 985 of 992 CT scans (99%) and in 488 of 496 patients (98%) in our study. In previous studies, the ICV was seen on 68% of barium enema studies and 72%–98% of abdominal CT scans (either conventional CT or CT colonography), with higher rates of localization corresponding to narrower collimation and reconstruction intervals (6,13,14,33).

The ICV varies in shape, size, fat content, CT attenuation, and location. At endoscopy, the ICV can be labial, with a slitlike opening (76% of cases), papillary (ie, dome shaped, 21% of cases), or lipomatous, with a substantial amount of fat deposition in the lips (3% of cases) (13). At barium enema examination, the valve is round or ovoid in 78% of patients versus triangular in 22% and smooth in 85% of patients versus lobulated in 15%. It has lips that are symmetric in 88% of patients and asymmetric in 12% of patients (14). It varies widely in height (range, 1–4 cm; average, 1.7 cm) and width (range, 1–6 cm; average, 2.8 cm) (14). In addition, the ICV may appear unusually large and polypoid due to ileal prolapse into the cecum (34–36). Although fat is usually present in the submucosa of the gastrointestinal tract, it is particularly common in the ICV and may give rise to fatty deposits or overall fatty infiltration and lipohyperplasia (15–18,33). In fact, an increase in the width of the ICV is almost always due to accumulation of adipose areolar tissue in the colonic aspect of the valve lips (37). This concentration of fat accounts for the low average attenuation of many ICVs. The ICV may be located medially (81%–90% of cases), laterally (1%–15% of cases), or posteriorly (3%–8% of cases) at the first (33%), second (65%), or third (2%) haustral fold and is associated with a fixed (64%) or mobile (36%) cecum (14,34). This variability may make it difficult to identify solely on the basis of location.

It is important to properly identify the ICV and thus prevent it from being labeled as a polyp; some argue that the ICV must be identified in every patient during CT colonography (7). If CAD findings of possible polyps are marked in one color and findings of ICVs marked with another color, the radiologist could inspect the ICV for abnormalities and differentiate it from other findings in the vicinity. Authors of most studies with two-dimensional CT suggest following the ileum into the colon to find the ICV (6,8,9,11,32). Until an automated ileum locator is successfully programmed, however, polyps and ICVs must be differentiated by means of their internal characteristics. The frequent presence of adipose tissue within the ICV can be used to differentiate it from a polyp (11,16). One CT colonographic study (16) proposed translucency renderings to highlight fat within the valve. Summers et al (6) offered an automatic way of identifying ICVs by means of volume and attenuation thresholds and stated that if thinner collimation is used (5 mm was used in the study), thresholds must be readjusted.

We have shown that volume and attenuation thresholds may be used to create a simple rule-based classifier to label ICVs when using CAD to interpret CT colonographic images while retaining accurate labels for most adenomatous polyps. Thresholds with 100% sensitivity and 83% specificity, which were established with a training set of cases, resulted in 97% sensitivity and 84% specificity when applied to the test set. If we assume that CAD helps detect all segmentable ICVs as polyps, these thresholds would reduce the false-positive rate by 1.6 false-positive finding per patient.

Although an increase in specificity is beneficial, it cannot come at the cost of decreased sensitivity. The object of screening examinations such as CT colonography is to detect polyps with malignant potential; it would be detrimental to lose all detections of a clinically important polyp (ie, a polyp that must be either followed up or removed). Application of the volume and attenuation thresholds established with the training set to the test set resulted in four of 138 polyps being mislabeled. These polyps, which were mislabeled as ICVs, would have to be correctly identified as polyps by the radiologist.

Our study differs from previous studies in several important ways, which can be illustrated by comparing it with the study by Summers et al (6), who evaluated the automatic recognition of ICVs. First, we used 1.25–2.5-mm collimation and a reconstruction interval of 1 mm; Summers et al used 5-mm collimation and a 3-mm reconstruction interval. Our testing and training data were from the same patient pool, whereas Summers et al used two different patient groups. We used oral contrast material to opacify stool and residual fluid because evidence supports this approach as providing a benefit for polyp detection (19,38). We used a three-dimensional polyp segmentation algorithm instead of a two-dimensional algorithm. In our experience, three-dimensional segmentation is better than two-dimensional segmentation because two-dimensional segmentation leads to undesirable stair-step artifacts. We based sensitivity on unique polyps rather than prone and supine polyp detections, which enabled one detection to be lost if another was retained on the other scan. With these refinements, we were able to attain both a high sensitivity and a higher specificity; Summers et al obtained a sensitivity of 100% and specificities of 61% and 33% in the training and test sets, respectively.

Opacification of residual fluid by oral contrast material offered additional challenges to our study because voxels with high-attenuation fluid or high attenuation due to volume averaging with opacified fluid could be included in the segmentation of a polyp or an ICV. To counteract the fluid voxels, all voxels with an attenuation of more than 276 HU were excluded from the final segmented ICV or polyp. To remove voxels that had volume averaging with opacified fluid or air, the edge of the initial segmentation was eroded away, leaving only voxels within—but not at the boundary of—the ICV or polyp. Despite these measures, the mean attenuation of polyps was –52 HU in the training set and –68 HU in the test set, which indicates that there was still residual partial volume averaging artifact.

A limitation of our study was that we did not include location data in our analysis of ICVs and polyps. We assumed that polyps in the cecum and the rest of the colon have a similar appearance at CT, although, to our knowledge, there are no data that confirm or dispute this. Although location information may help in the diagnosis of an ICV by an ICV detector, our data about true polyps were insufficient to support an analysis including this variable. In clinical practice, an algorithm to eliminate ICVs would only apply to findings in the cecum.

Another potential limitation of our study is that our algorithm did not detect that some ICVs are flat or have thin lips. It is possible that the computation of additional features (eg, flatness) will be of value in future research.

In summary, we have developed a scheme to use volume and attenuation data to label ICVs while retaining clinically important polyps. Our thresholds were set by using a training set of cases, in which 100% sensitivity and 83% specificity were obtained, and applied to a test set of cases, in which 97% sensitivity and 84% specificity were obtained. The incorporation of volume and attenuation data into a CAD algorithm for CT colonography may enable the identification and inspection of the ICV, reduce the number of false-positive findings, and increase the efficiency of the examination—all of which remain to be verified.


   ADVANCES IN KNOWLEDGE
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
References
 


   ACKNOWLEDGMENTS
 
We thank William Schindler, DO, Naval Medical Center, San Diego, for providing CT colonography and supporting data; Andrew Dwyer, MD, for critical review of the manuscript; and Daniel Brickman, BS, and Meghan Miller, BS, for creating a table of CT image coordinates for polyps on the basis of optical colonoscopy data.


   FOOTNOTES
 

Abbreviations: CAD = computer-aided detection • CI = confidence interval • ICV = ileocecal valve

See Materials and Methods for pertinent disclosures.

Author contributions: Guarantor of integrity of entire study, R.M.S.; 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.D.O., R.M.S.; clinical studies, P.J.P., J.R.C.; experimental studies, S.D.O., R.M.S., J.Y.; statistical analysis, S.D.O., R.M.S., J.Y.; and manuscript editing, all authors


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 ADVANCES IN KNOWLEDGE
 References
 

  1. Summers RM, Beaulieu CF, Pusanik LM, et al. Automated polyp detector for CT colonography: feasibility study. Radiology 2000;216:284–290.[Abstract/Free Full Text]
  2. Summers RM, Jerebko AK, Franaszek M, Malley JD, Johnson CD. Colonic polyps: complementary role of computer-aided detection in CT colonography. Radiology 2002;225:391–399.[Abstract/Free Full Text]
  3. Summers RM, Johnson CD, Pusanik LM, Malley JD, Youssef AM, Reed JE. Automated polyp detection at CT colonography: feasibility assessment in a human population. Radiology 2001;219:51–59.[Abstract/Free Full Text]
  4. Yoshida H, Masutani Y, MacEneaney P, Rubin DT, Dachman AH. Computerized detection of colonic polyps at CT colonography on the basis of volumetric features: pilot study. Radiology 2002;222:327–336.[Abstract/Free Full Text]
  5. Paik DS, Beaulieu CF, Rubin GD, et al. Surface normal overlap: a computer-aided detection algorithm with application to colonic polyps and lung nodules in helical CT. IEEE Trans Med Imaging 2004;23:661–675.[CrossRef][Medline]
  6. Summers RM, Yao J, Johnson CD. CT colonography with computer-aided detection: automated recognition of ileocecal valve to reduce number of false-positive detections. Radiology 2004;233:266–272.[Abstract/Free Full Text]
  7. Johnson CD, Dachman AH. CT colonography: the next colon screening examination? Radiology 2000;216:331–341.[Abstract/Free Full Text]
  8. Macari M, Megibow AJ. Pitfalls of using three-dimensional CT colonography with two-dimensional imaging correlation. AJR Am J Roentgenol 2001;176:137–143.[Free Full Text]
  9. Fenlon HM. CT colonography: pitfalls and interpretation. Abdom Imaging 2002;27:284–291.[Medline]
  10. Fenlon HM, Clarke PD, Ferrucci JT. Virtual colonoscopy: imaging features with colonoscopic correlation. AJR Am J Roentgenol 1998;170:1303–1309.[Free Full Text]
  11. Macari M, Bini EJ, Jacobs SL, Lange N, Lui YW. Filling defects at CT colonography: pseudo- and diminutive lesions (the good), polyps (the bad), flat lesions, masses, and carcinomas (the ugly). RadioGraphics 2003;23:1073–1091.[Abstract/Free Full Text]
  12. Summers RM. Challenges for computer-aided diagnosis for CT colonography. Abdom Imaging 2002;27:268–274.[Medline]
  13. Regge D, Gallo TM, Nieddu G, et al. Ileocecal valve imaging on computed tomographic colonography. Abdom Imaging 2005;30:20–25.[CrossRef][Medline]
  14. El-Amin LC, Levine MS, Rubesin SE, Shah JN, Kochman ML, Laufer I. Ileocecal valve: spectrum of normal findings at double-contrast barium enema examination. Radiology 2003;227:52–58.[Abstract/Free Full Text]
  15. Tawfik OW, McGregor DH. Lipohyperplasia of the ileocecal valve. Am J Gastroenterol 1992;87:82–87.[Medline]
  16. Pickhardt PJ. Translucency rendering in 3D endoluminal CT colonography: a useful tool for increasing polyp specificity and decreasing interpretation time. AJR Am J Roentgenol 2004;183:429–436.[Free Full Text]
  17. Hulten J. Lipomatosis of the ileocaecal valve. Acta Chir Scand 1965;129:104–111.[Medline]
  18. Fultz PJ, Hampton WR, Skucas J, Sickel JZ. Differential diagnosis of fat-containing lesions with abdominal and pelvic CT. RadioGraphics 1993;13:1265–1280.[Abstract/Free Full Text]
  19. Pickhardt PJ, Choi JR, Hwang I, et al. Computed tomographic virtual colonoscopy to screen for colorectal neoplasia in asymptomatic adults. N Engl J Med 2003;349:2191–2200.[Abstract/Free Full Text]
  20. Summers RM, Yao J, Pickhardt PJ, et al. Computed tomographic virtual colonoscopy computer-aided polyp detection in a screening population. Gastroenterology 2005;129:1832–1844.[CrossRef]
  21. Yao J, Miller MT, Summers RM. Automatic segmentation and detection of colonic polyps in CT colonography based on knowledge-guided deformable models. In: Clough A, Amini A, eds. SPIE medical imaging. San Diego, Calif: International Society for Optical Engineering, 2003; 370–380.
  22. Yao J, Miller M, Franaszek M, Summers RM. Colonic polyp segmentation in CT colonography-based on fuzzy clustering and deformable models. IEEE Trans Med Imaging 2004;23:1344–1352.[CrossRef][Medline]
  23. Summers RM, Franaszek M, Miller MT, Pickhardt PJ, Choi JR, Schindler WR. Computer-aided detection of polyps on oral contrast-enhanced CT colonography. AJR Am J Roentgenol 2005;184:105–108.[Free Full Text]
  24. Preacher KJ, Briggs NE. Calculation for Fisher's exact test. Available at: http://www.unc.edu/~preacher/fisher/fisher.htm.
  25. Lowry R. Clinical calculator 1: from an observed sample—estimates of population prevalence, sensitivity, specificity, predictive values, and likelihood ratios. Available at: http://faculty.vassar.edu/lowry/clin1.html.
  26. Freer TW, Ulissey MJ. Screening mammography with computer-aided detection: prospective study of 12,860 patients in a community breast center. Radiology 2001;220:781–786.[Abstract/Free Full Text]
  27. Irwig L, Houssami N, van Vliet C. New technologies in screening for breast cancer: a systematic review of their accuracy. Br J Cancer 2004;90:2118–2122.[Medline]
  28. Hadjiiski L, Chan HP, Sahiner B, et al. Improvement in radiologists' characterization of malignant and benign breast masses on serial mammograms with computer-aided diagnosis: an ROC study. Radiology 2004;233:255–265.[Abstract/Free Full Text]
  29. Warren Burhenne LJ, Wood SA, D'Orsi CJ, et al. Potential contribution of computer-aided detection to the sensitivity of screening mammography. Radiology 2000;215:554–562.[Abstract/Free Full Text]
  30. Kakeda S, Moriya J, Sato H, et al. Improved detection of lung nodules on chest radiographs using a commercial computer-aided diagnosis system. AJR Am J Roentgenol 2004;182:505–510.[Abstract/Free Full Text]
  31. Awai K, Murao K, Ozawa A, et al. Pulmonary nodules at chest CT: effect of computer-aided diagnosis on radiologists' detection performance. Radiology 2004;230:347–352.[Abstract/Free Full Text]
  32. Hara AK, Johnson CD, Reed JE. Colorectal lesions: evaluation with CT colography. RadioGraphics 1997;17:1157–1167, discussion 1167–1168.[Abstract]
  33. Silverman PM, Kelvin FM, Baker ME, Cooper C. Computed tomography of the ileocecal region. Comput Med Imaging Graph 1988;12:293–303.[CrossRef][Medline]
  34. Hinkel CL. Roentgenological examination and evaluation of the ileocecal valve. Am J Roentgenol Radium Ther Nucl Med 1952;68:171–182.[Medline]
  35. Hawley C, Mithoefer J. Ileal prolapse. Radiology 1950;54:380–385.[Medline]
  36. Perkel LL, Troast L. Ileocecal valve prolapse simulating cecal polyp. Am J Gastroenterol 1955;23:103–108.[Medline]
  37. Fleischner FG, Bernstein C. Roentgen-anatomical studies of the normal ileocecal valve. Radiology 1950;54:43–58.[Medline]
  38. Pineau BC, Paskett ED, Chen GJ, et al. Virtual colonoscopy using oral contrast compared with colonoscopy for the detection of patients with colorectal polyps. Gastroenterology 2003;125:304–310.[CrossRef]



This article has been cited by other articles:


Home page
 Am. J. Roentgenol.Home page
S. A. Taylor, J. Brittenden, J. Lenton, H. Lambie, A. Goldstone, P. N. Wylie, D. Tolan, D. Burling, L. Honeyfield, P. Bassett, et al.
Influence of Computer-Aided Detection False-Positives on Reader Performance and Diagnostic Confidence for CT Colonography
Am. J. Roentgenol., June 1, 2009; 192(6): 1682 - 1689.
[Abstract] [Full Text] [PDF]

Home page
 Am. J. Roentgenol.Home page
J. G. Fletcher, F. Booya, R. M. Summers, D. Roy, L. Guendel, B. Schmidt, C. H. McCollough, and J. L. Fidler
Comparative Performance of Two Polyp Detection Systems on CT Colonography
Am. J. Roentgenol., August 1, 2007; 189(2): 277 - 282.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Right arrow All Versions of this Article:
2412051223v1
241/2/426    most recent
Right arrow Submit a response
Right arrow Alert me when this article is cited
Right arrow Alert me when eLetters are posted
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by O'Connor, S. D.
Right arrow Articles by Choi, J. R.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by O'Connor, S. D.
Right arrow Articles by Choi, J. R.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
RADIOLOGY RADIOGRAPHICS RSNA JOURNALS ONLINE