| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Technical Innovation |
1 Radiology Department, National Institutes of Health, 10 Center Dr., Bldg. 10,
Rm. 1C660, MSC 1182, Bethesda, MD 20892-1182.
2 National Naval Medical Center, Bethesda, MD 20892.
3 Uniformed Services University of the Health Sciences, Bethesda, MD
20814.
4 Present address: Department of Radiology, University of Wisconsin Medical
School, Madison, WI 53792.
5 Walter Reed Army Medical Center, Washington, DC 20307.
6 Naval Medical Center San Diego, San Diego, CA 92134.
Received February 12, 2004;
accepted after revision April 28, 2004.
Address correspondence to R. M. Summers
(rms{at}nih.gov).
Introduction
 Top Introduction Materials and MethodsResultsDiscussionReferences |
|---|
|
TopIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
At all institutions, the patient inclusion criteria specified average-risk asymptomatic adults between 50 and 79 years old who were referred for colorectal cancer screening and asymptomatic patients 40–79 years old with a first-degree relative with a history of colorectal cancer. The protocol was approved by the institutional review boards at all three institutions, and informed consent was obtained from all patients.
CT Protocol
Patient preparation.—The bowel preparation included 500 mL
of dilute CT barium solution 2.1% by weight (Scan C, Lafayette
Pharmaceuticals), 120 mL of diatrizoate meglumine and diatrizoate sodium
solution (Gastroview, Mallinckrodt; or Gastrografin, Bracco Diagnostics), and
90 mL of sodium phosphate (24-hr Fleet 1 preparation, Fleet Pharmaceuticals)
in divided doses. The patients also followed a clear liquid diet and took two
bisacodyl tablets (Dulcolax, Boehringer Ingelheim).
Scanning.—Colonic distention was achieved by patient-controlled rectal insufflation of room air to achieve maximum colonic distention. Each patient was scanned supine and prone in a 4-MDCT scanner (LightSpeed Plus, GE Healthcare) or an 8-MDCT scanner (LightSpeed Ultra, GE Healthcare) with 4 x 2.5 mm or 8 x 1.25 mm detector configuration, respectively; 2.5- or 1.25-mm collimation, respectively; and 1-mm reconstruction interval. Other CT parameters were a table speed of 15 mm per second, 120 kVp, and an effective tube current of 100 mAs. For the purposes of this technical report, only one of the two views (nine supine, eight prone) was analyzed.
Bowel segmentation method.—We developed a colon segmentation method using a region-growing algorithm that, like a porpoise, "jumps" from air to fluid and back again until all portions of the colon are identified. The procedure is shown schematically in Figure 1. It consists of the following main steps: first, region-growing segmentation with capability to traverse smoothly from air to opacified fluid and vice versa; second, labeling of air and fluid regions and calculation of mean and SD of total colonic fluid CT attenuation; third, identification and labeling of air–fluid boundaries; and, fourth, second segmentation of fluid-filled segments to correct for possible leakage from the first segmentation.
|
The first segmentation (step 1) is performed with predefined thresholds for
air (Tair) of –800 H and for fluid (Tfluid) of 276
H. Voxels with intensity below Tair are labeled air; those with
intensity higher than Tfluid, as fluid; and all others, as colonic
wall. The relatively low value of Tfluid ensures that fluid-filled
colonic segments are not missed. On the other hand, low Tfluid may
cause leakage to adjacent structures such as small bowel. For this undesired
effect to be minimized, transitions between fluid and air are allowed only
under very restrictive conditions: the thickness of the air–fluid
boundary cannot exceed 2 voxels, and an air region cannot be below a fluid
region (the fluid is dependent with respect to gravity). After computation of
the mean fluid intensity,
, and the SD of fluid
intensity, 
fluid (step 2), air–fluid boundaries are
identified (step 3) on the basis of the geometric conditions such as maximum
allowed thickness and maximum acceptable slope (air–fluid boundaries are
flat). Then (step 4), a modified fluid threshold is defined as follows:
 |
Computer-Aided Polyp Detection Method
We modified our existing CAD system to detect polyps submerged in opacified
colonic fluid [4]. Once the
segmentation is completed, all voxel labels fall into five different classes:
air, fluid, air–fluid boundaries, colonic wall, and other. The colonic
wall is constructed by a modified isosurface procedure. A surface is built
between the following pairs of voxels: air–wall, fluid–wall, and
air–fluid boundary and wall; all other combinations of labels are
ignored. Depending on the particular pair of labels, the corresponding
threshold is selected adaptively: Tair for air–wall and
Tfluid' for fluid–wall; for air–fluid boundary and
wall, we place the corresponding vertex in the midpoint between the
air–fluid boundary and wall voxel. Every vertex on the surface keeps
information about the pair of labels that contributed to its creation.
Once construction of the colonic wall is complete, the curvature for every vertex is determined on the basis of a convolution method and gradient calculations. To get a properly defined curvature for polyps located both in air and under fluid, we have to reverse the sign for the principal components of curvature calculated for vertices under fluid. Once the curvature for every vertex is found, a process of clustering starts: All mutually connected vertices that pass certain predefined curvature tests (elliptic type and mean value between –4 and –0.2 cm–1) are selected as possible polyp candidates. Parameters of the segmentation again depend on whether a candidate is located in air or in fluid.
Assessment of the Method
For each case, we assessed visually whether segmentation of the colon was
complete or incomplete. For this assessment, small missed air pockets were not
considered significant. We assessed the amount of small-bowel leakage (defined
as when parts of the small bowel filled with oral contrast material or air and
inappropriately are considered to be parts of the colon) as none, small,
medium, or extensive by visual assessment while scrolling through the stack of
CT colonography images; on these images, the segmented air and fluid within
the colon were color-coded to distinguish them from unsegmented air and fluid.
We have proved the efficacy of the new algorithm by reporting the number of
polyps submerged in fluid that were found using CAD. We show a color-coded
surface reconstruction of a computer-aided detection on a known polyp
submerged under fluid.
|
TopIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
|
|
The colonic segmentation was complete in all cases. The median number of seeds needed to segment the colon was one (range, one to four; mean, 1.6 ± 1.0 [SD]). More than one seed was needed only if the colon was collapsed, resulting in two or more disconnected colonic segments. The amount of leakage into small bowel was none in 10 patients, small in five, medium in one, and extensive in one.
Six of the 22 polyps were on the air–fluid boundaries (part covered by fluid, part by air). Nineteen (86%) of the 22 polyps were detected by CAD. Of the three false-negatives, two were 0.5 cm and one, a hyperplastic polyp on an air–fluid boundary, was 0.7 cm. Figures 3A, 3B, and 3C shows an example of a true-positive detection of a polyp submerged in opacified fluid.
|
|
|
|
TopIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
For CT colonography, the goal is to design an algorithm for visualization of the colon wall. To this end, various methods of fluid subtraction, electronic cleansing, or digital bowel cleansing have been proposed [5–7]. For CAD, the focus is the quantitative analysis of colonic wall features for polyp detection. In both cases, the fundamental difficulty lies in adequately distinguishing the range of tagged remnants from other parts of the colon.
On the basis of our finding that intrapatient variation in fluid attenuation was low [8], we determined that we could use a simple adaptive thresholding algorithm to identify polyps submerged in opacified fluid and wall structure. Wyatt et al. [7] administered iodinated oral contrast material to patients a few hours before the examination and reported variability of fluid density along the length of the colon by "as much as 10%" in a few subjects. They concluded that this variation prevents the use of a constant threshold. We found a smaller degree of variation [8]. In addition, our CAD scheme was relatively tolerant of this variation because the constant threshold is only the first stage in a multistage scheme.
Exclusion of small bowel from further processing is important to reduce undesirable false-positive detections. We found that in most of the cases (15/17, 88%), there was minimal or no leakage of the colonic segmentation into the small bowel. Our software provides the additional capability to exclude manually most of the small bowel in the few cases in which leakage is large.
In actual practice, the CAD algorithm described herein is applied both to the air-filled and to the fluid-filled portions of the colon. In this way, polyps surrounded by either air or fluid may be detected. We have shown that polyps at the air–fluid boundaries—that is, with one portion touching air and another touching fluid—also can be detected with this scheme.
Our CAD algorithm could be enhanced by setting an initial threshold to classify most of the fluid and then use local analysis to refine the classification. Future work will address optimization of the algorithm and reporting overall sensitivity and false-positive rates for larger numbers of patients.
In summary, we have presented a CAD system that can detect polyps submerged in opacified colonic fluid.
Acknowledgments
We thank Shawn Albert for data management and Andrew Dwyer for manuscript
review.
|
TopIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  C. Robinson, S. Halligan, S. A. Taylor, S. Mallett, and D. G. Altman CT Colonography: A Systematic Review of Standard of Reporting for Studies of Computer-aided Detection Radiology, February 1, 2008; 246(2): 426 - 433. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. E. Baker, L. Bogoni, N. A. Obuchowski, C. Dass, R. M. Kendzierski, E. M. Remer, D. M. Einstein, P. Cathier, A. Jerebko, S. Lakare, et al. Computer-aided Detection of Colorectal Polyps: Can It Improve Sensitivity of Less-Experienced Readers? Preliminary Findings Radiology, October 1, 2007; 245(1): 140 - 149. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. T. Johnson, J. G. Fletcher, and C. D. Johnson Computer-Aided Detection (CAD) Using 360{degrees} Virtual Dissection: Can CAD in a First Reviewer Paradigm Be a Reliable Substitute for Primary 2D or 3D Search? Am. J. Roentgenol., October 1, 2007; 189(4): W172 - W176. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
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]
|
|
|
|
|
|
P. J. Pickhardt Screening CT Colonography: How I Do It Am. J. Roentgenol., August 1, 2007; 189(2): 290 - 298. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Huang, D. A. Roy, R. M. Summers, M. Franaszek, N. Petrick, J. R. Choi, and P. J. Pickhardt Teniae Coli-based Circumferential Localization System for CT Colonography: Feasibility Study Radiology, May 1, 2007; 243(2): 551 - 560. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. D. O'Connor, R. M. Summers, J. Yao, P. J. Pickhardt, and J. R. Choi CT Colonography with Computer-aided Polyp Detection: Volume and Attenuation Thresholds to Reduce False-Positive Findings Owing to the Ileocecal Valve Radiology, November 1, 2006; 241(2): 426 - 432. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. Shi, P. Schraedley-Desmond, S. Napel, E. W. Olcott, R. B. Jeffrey Jr, J. Yee, M. E. Zalis, D. Margolis, D. S. Paik, A. J. Sherbondy, et al. CT Colonography: Influence of 3D Viewing and Polyp Candidate Features on Interpretation with Computer-aided Detection. Radiology, June 1, 2006; 239(3): 768 - 776. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. P. Mulhall, G. R. Veerappan, and J. L. Jackson Meta-Analysis: Computed Tomographic Colonography Ann Intern Med, April 19, 2005; 142(8): 635 - 650. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
