| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Radiology, Mayo Clinic, 200 First St. SW, Rochester, MN
55905.
2 Department of Physiology and Biomedical Engineering, Mayo Clinic, Rochester,
MN.
3 Facilities and Systems Support Services, Mayo Clinic, Rochester, MN.
Received August 23, 2005;
accepted after revision December 7, 2005.
Address correspondence to C. D. Johnson.
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
MATERIALS AND METHODS. A phantom containing 6-mm flat polyps and three types of simulated stool (homogeneous, moderately heterogeneous, and severely heterogeneous) mixed with barium was created, scanned, and tested using three stool subtraction algorithms but no cathartic. Thirty patients with suspected colorectal polyps were studied using stool tagging to determine which was the most effective stool subtraction algorithm. Colonoscopy was the reference standard. Examinations were evaluated blindly using the unsubtracted and 6 weeks later both the unsubtracted and subtracted data sets.
RESULTS. A threshold of 200 H and expansion and convolution
techniques were the most effective tools for subtracting stool and minimizing
artifacts. When applied to the human population, sensitivities using the
unsubtracted data sets were 90% (18/20) and 68% (26/38) for polyps 
1 cm
and 5 mm, respectively. Specificities were 100% (4/4) and 75% (3/4) for
polyps 1 cm and 5 mm. For the stool-subtracted data sets,
sensitivities were 90% (18/20) and 71% (27/38) for polyps 1 cm and 5
mm. Per patient sensitivities were 88% (15/17) and 77% (20/26) for 1 cm
and 5 mm polyps. Specificities were 100% (4/4) for large polyps and 25%
(1/4) for smaller polyps.
CONCLUSION. Image processing tools combining thresholding, expansion, and convolution were the most useful for stool subtraction. Laxative-free colon examinations using barium for stool labeling can be performed at CT colonography with or without stool subtraction with high accuracy. Further study is warranted.
Keywords: colonography • CT • gastrointestinal imaging • screening • virtual colonoscopy
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Although colorectal cancer is a serious disease, many patients are often reluctant to undergo screening because of the required bowel preparation. Cathartic preparation can be uncomfortable and inconvenient. A large study evaluating patient perceptions and preferences at CT colonography and colonoscopy found that 72% of patients preferred CT colonography over colonoscopy, and a significant number of patients would have the examination more frequently if bowel preparation was not required [5]. Ideally, bowel preparation can be eliminated so that more patients undergo screening. It is possible to label the stool with a contrast agent [6] and theoretically possible to electronically subtract stool, saving patients the need to undergo laxative preparation.
The aim of this feasibility study was to develop a method to subtract barium-labeled stool using a phantom and to evaluate the performance of this technique in a pilot human population.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
|
Phantom Study
Simulated stool creation—A phantom was created to test stool
subtraction algorithms in a controlled environment with one type of polyp
morphology (flat morphology, 6-mm diameter, 3-mm height) at four distinct
locations on a haustral fold. The phantom provided humanlike conditions for
testing the stool subtraction algorithms but avoided the unpredictable
variability of a patient population. Flat polyps were selected because they
are the most difficult to detect, and 6 mm is an important clinical threshold
for polyp size (Figs. 1A and
1B).
Artificial stool was created using magnesium aluminum silicate (Veegum, R. T. Vanderbilt Company). This material is used as simulated stool for nuclear medicine proctographic studies. Water (50 mL) and varying amounts of barium (0.3-2.5 g) were added to magnesium aluminum ciliate (30 mL) and scanned using CT. Measurements of mean attenuation were made of each mixture. The mixture with a mean attenuation closest to 350 H was chosen for the experiment. (See the Results section, Simulated Stool Creation.)
Three heterogeneities (homogeneous, moderately heterogeneous, and severely heterogeneous) of stool were created. Homogeneous stool was created by mixing 30 mL of magnesium aluminum silicate, 0.45 g of barium, and 50 mL of water (derived from the measurements described previously). Heterogeneous stool was made with magnesium aluminum silicate, barium, water (as listed previously), and chopped carrots. Mixtures were scanned at CT and attenuation values measured. It was determined that 20 and 30 mL of chopped carrots were needed for moderate and severe heterogeneity, respectively. Barium was increased to 0.9 g for severely heterogeneous stool to maintain an attenuation of 350 H.
Phantom creation and scanning—A glass colon phantom with haustral folds and bends was created and served as the mold for a Silastic (liquid silicone rubber C6-515, parts A and B, Dow Corning) colon phantom [7]. One hundred grams of Silastic was mixed with 75 mL of silicone fluid, poured through the glass mold several times, and cured at 150°C. The Silastic phantom was then peeled away from the two-piece glass mold. The various segments of the colon phantom were joined by shrink-wrapping them together to create an anatomically correct model. The wall of the phantom measured approximately 200 H.
Sixteen flat polyps (6-mm long and 3-mm high) were glued into the phantom at different locations (on the wall, on the tip of the fold, and on the fold). Each of the three types of simulated stool were individually injected into the phantom to cover the polyps. The phantom was sealed (watertight) and placed in a life-sized water bath and scanned on an 8-MDCT scanner (LightSpeed Ultra, GE Healthcare) using the following technique: 1.25-mm slice thickness, 1.25-mm reconstructive intervals, 120 kVp, 140 mA, 0.5-second rotation time, and 40-cm field of view. This protocol was repeated for each type of stool.
Phantom data analysis—The phantom CT data were networked to a computer running specialized software (Analyze 5.0, Mayo Foundation). The stool subtraction algorithms were evaluated in series after feedback from the prior analysis. For each algorithm, the 16 flat polyps in the phantom were assessed using the unsubtracted and stool-subtracted (combination of unsubtracted and subtracted) data sets. The percentage of stool subtracted, the number of polyps found, and various artifacts from the subtraction algorithms were recorded using a severity index of 0-3 (0 = none, 3 = severe). Artifacts that were observed included halos, islands, rough borders, partial and complete polyp subtraction, and fold subtraction. Halos were defined as rings of barium along the periphery of the stool after subtraction. Islands were clumps of simulated stool surrounded by air on an axial image. Rough borders and fold subtraction referred to irregularly surfaced walls and truncation of haustral folds, respectively.
|
|
|
The second algorithm used a threshold of 200 H plus dilation of the area of the colon containing stool to remove halo artifacts. The dilation step adds voxels to the portion of the stool region that is in contact with air. To avoid tissue distortion, stool voxels adjacent to soft tissue are not dilated.
The third algorithm used a method that was previously described in the study by Carston et al. [8]. This method requires knowledge of the point spread function (PSF) of the scanner. An estimate of the PSF was determined empirically using a standard phantom object with small metal beads. The continuous PSF was estimated to be gaussian and to have an SD of 0.85 mm. A binary version of the PSF was constructed using a 3 x 3 x 3 element with the eight corners removed—that is, an 18-connected neighborhood. The algorithm begins by attempting to classify voxels as stool, tissue, or air on the basis of intensity and gradient values. Voxels that are not classified as any tissue type are classified as unknown. An iterative classifier then attempts to classify each unknown voxel by examining its 26 neighbors and applying various decision rules. The stool, air, and tissue masks are then dilated using the binary PSF structuring element. For any voxels in which the expanded stool and tissue region intersect, we simply assume that the voxels belong to the tissue mask, which helps avoid polyp distortion. The expanded stool mask is then convolved with the continuous PSF, resulting in an estimate of the percentage of stool at each voxel. The tagged stool is subtracted from the image by reducing the voxel intensities by an amount proportional to the estimated percentage of stool.
Human Study
The study was approved by the institutional review board, and informed
consent was obtained from all patients. Thirty patients with a known or
suspected colon polyp or cancer were recruited for the study. Patients who
were mentally disabled, prisoners, and pregnant women were not included in
this study. All colons were prepared with barium tagging of stool before
examination as detailed in the Phantom Study portion of this section. All
patients in the study underwent CT colonography and, on a subsequent day,
complete colonoscopy that required a standard bowel preparation. Preparation
was satisfactory in all patients. Colonoscopy served as the reference
standard. Stool tagging was considered adequate in all patients.
CT Colonography
All patients received glucagon, 1 mg subcutaneously, unless contraindicated
or refused by the patient, 10 minutes before the CT examination. Patients were
placed in either lateral decubitus position for enema tip insertion and slow
manual insufflation of approximately 2 L of carbon dioxide until the patient
verbally indicated air administration had reached maximal tolerance. Both
supine and prone data acquisitions were obtained, and additional carbon
dioxide was added as tolerated by the patient before scanning in the prone
position to compensate for any lost during position changes.
All examinations were performed using a Light-Speed Ultra 8-MDCT scanner (GE Healthcare). After colon insufflation, a breath-hold anteroposterior scout image was obtained before each acquisition to assess luminal distention and to prescribe axial slices through the entire large bowel. Images were acquired using 2.5-mm collimation, table speed of 13.5 mm/s, 1.25-mm reconstruction intervals, a matrix of 512 x 512, a field-of-view to fit, 140 mA, 120 kVp, 0.5-second rotation time, and a standard reconstruction algorithm. The entire abdomen and pelvis were scanned in a single breath-hold. The image data were retrospectively reconstructed at 1.25-mm slice thickness at 1.25-mm intervals. Only the 1.25-mm slice thickness data were used for this study.
|
|
|
|
|
|
|
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Algorithm analysis—The effect of threshold values (100, 150, and 200 H) causing polyp subtraction is shown in Figure 4. As threshold values increased, polyp subtraction decreased. No polyps were subtracted at a threshold of 200 H. Figure 5A summarizes stool subtraction artifacts generated using algorithm 1 at a threshold of 200 H. None of the polyps or folds were subtracted, but severe levels of halos, islands, and rough luminal borders were present (Figs. 6A and 6B). Figure 5B summarizes the findings using stool subtraction algorithm 2, which incorporated a threshold of 200 H and expansion of the stool mask. Halos were eliminated and other artifacts reduced from severe to moderate levels. Mild (partial) polyp subtraction did occur (Figs. 7A and 7B). Figure 5C depicts the results of stool subtraction algorithm 3, which used a threshold of 200 H, expansion, and convolution. Mild halos, islands, and rough borders predominated without polyp subtraction. Stool subtraction algorithm 3 was used for the human study (Figs. 8A, 8B, 9A, and 9B).
Human Study
When viewing the unsubtracted stool data sets, the per polyp sensitivities
were 90% and 68% for polyps 1 cm and 5 mm in diameter, respectively.
Per patient sensitivities were 88% and 77% for polyps 1 cm and 5 mm.
Specificities were 100% for larger polyps and 75% for polyps 5 mm (Table
1).
Using stool subtraction algorithm 3, the per polyp sensitivities were 90%
for polyps 1 cm and 71% for polyps 5 mm. The per patient
sensitivities for the subtracted data were 88% for large polyps and 77% for
smaller polyps. Specificities for the subtracted stool examinations were 100%
for large polyps and 25% for polyps 5 mm (Table 1).
Causes of large polyp (> 1 cm) error among the unsubtracted data sets
included one perceptive error (polyps could be seen in retrospect) and one
technical error (invisible in retrospect). Among the 12 missed polyps 5
mm, half were missed because of perceptive errors and the other half were
technical errors. There were two large polyp errors among the subtracted data
sets—one each of a perceptive and technical cause. Of missed polyps
5 mm, 8 of 11 were perceptive errors and three were technical. Average
interpretation times (minutes:seconds) for the subtracted and unsubtracted
interpretations were 14:33 (range, 9:17-30:00) and 12:04 (range, 8:00-20:42),
respectively.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Simulated stool was produced with imaging characteristics that mimic human stool. Stool subtraction algorithms can be evaluated using a novel colon phantom containing simulated stool. Image processing methods, including thresholding, expansion, and convolution, are effective but not perfect in electronic stool subtraction. Both unsubtracted and stool-subtracted data sets are effective for the detection of clinically important polyps.
Human stool has a high range of attenuation throughout the colon. This range increases in the distal colon. This broad range of stool attenuation and heterogeneity may contribute to the difficulty in achieving complete stool subtraction. Methods that increase stool homogeneity and maintain a high attenuation may translate into improved stool subtraction.
Creation of a realistic colon phantom for testing stool subtraction algorithms proved to be a challenge. A glass colon phantom that had previously been reported and tested was not usable because of its high-attenuation walls [9]. Silastic proved to be a material that could realistically represent the colon wall, haustrations, and flexures. The wall of the Silastic phantom was also of abnormally high attenuation compared with the normal colon but was of markedly lower attenuation than that of the glass phantom. As a result, wall erosion caused by the subtraction process could not be assessed. Not all types of polyps and sizes were tested; however, the type of polyp most challenging to detect clinically (6-mm diameter, flat) was assessed using a clinically realistic CT colonography technique.
Stool subtraction is difficult and requires complex image-processing techniques to remove unwanted stool without eroding normal colon structures and polyps. Stool subtraction algorithm 1 used only a threshold technique (Fig. 4). A threshold of 200 H was chosen because no polyps were subtracted at this level. As expected, severe subtraction artifacts were produced, including halos, islands, and rough (irregularly surfaced) borders. Stool subtraction algorithm 2 combined expansion and thresholding techniques and resulted in elimination of halo artifacts at the expense of mild polyp subtraction. Other artifacts (islands, rough borders, and so forth) persisted, but at lower, moderate levels. Stool subtraction algorithm 3 added convolution (derived from the PSF of the CT scanner) to thresholding and expansion. All subtraction artifacts were graded as mild, and no polyp subtraction occurred. As a result, stool subtraction algorithm 3 was used to evaluate the patient population.
All patients had stool tagging with barium but no laxative preparation or dietary restrictions as described in the phantom study before examination. Patients tolerated the barium tagging in a similar manner to the prepared colon.
The detection of large ( 1 cm) polyps using unsubtracted and
stool-subtracted data sets was 90% (18/20). The similar performance of both
interpretation methods is somewhat surprising because stool can be an
important distraction to the reviewer, even when optimally labeled. In
addition, subtracting stool should provide the radiologist with a quicker and
more effective examination than unsubtracted data sets. A modest reduction in
interpretation times occurred when using subtracted data sets. Average
interpretation times for the subtracted interpretations were approximately 2.5
minutes shorter than those using unsubtracted data sets. Subtraction and
faster interpretation times may have contributed to the lower specificity
because truncated folds can simulate small polyps. It is expected that as
stool subtraction techniques improve, the interpretation times can be
shortened.
Little difference in performance existed between subtracted and
unsubtracted interpretations for the detection of polyps 5 mm. The
sensitivities for the detection of polyps 5 mm using unsubtracted and
stool-subtracted data sets were 68% (26/38) and 71% (27/38), respectively.
Perceptive errors accounted for 50-73% of errors. These data are consistent with reports of the prepared colon [10] in which one third to two thirds of polyp errors were technical (not visible in retrospect) in origin. Most of the false-positive findings in our study were due to suboptimally labeled stool that had a polypoid morphology or truncated folds in subtracted data sets. Improving stool labeling should reduce technical errors in the future. None of the polyps was electronically subtracted.
Interpretations were made by a single experienced radiologist. Future comparison of findings with reviewers having less experience would be of interest. Our study findings are consistent with those of an earlier study by Sheppard et al. [11], who used an unprepared colon phantom to investigate stool subtraction on CT colonography. In that study, the authors used an animal colon with simulated polyps and simulated stool containing contrast material and peanut butter. They found the sensitivity and specificity for polyp detection to be 94% and 80%, respectively, for polyps 3 mm or larger.
Callstrom et al. [6] studied
the oral administration of contrast material in humans using different amounts
of barium given by mouth over a 24- and a 48-hour time period. Those authors
found that 18 g of barium administered over 48 hours in seven divided doses
resulted in optimal stool labeling throughout the colon. Polyps 1 cm in
diameter were detected with a sensitivity of 80-100% without subtraction. This
administration algorithm was used in our study— each patient ingested
20.5 g of barium—and confirms their findings.
Lefere et al. [12] reported a method of fecal tagging without cathartic cleansing. This method used a low-residue diet, hydration control, and various concentrations of barium. In another publication, the same group also reported successful tagging using lower (50 mL) volumes of barium [13]. Polyp detection results were not included.
Recent studies of the unprepared colon using iodinated contrast materials
for stool tagging by Zalis et al.
[14,
15] found the sensitivity for
the detection of polyps 1 cm was 60-70%. Our study differs in that only
barium was used as a tagging agent and a different subtraction algorithm was
used.
Iannaccone et al. [16]
reported a patient study using 200 mL of oral water-soluble iodinated contrast
material over a 2-day period without stool subtraction. The average
sensitivity and specificity for polyps 8 mm were 95% and 92%,
respectively. Although those results are impressive, water-soluble contrast
material can result in cramping and diarrhea in many patients. Our study used
only barium and was not associated with complaints of cramping or
diarrhea.
Lakare et al. [17] reported the results of a stool subtraction program in the unprepared colon. Patients ingested a soft diet 24 hours before examination and drank three bottles (amount not specified) of an unspecified "density-enhancing" fluid. Segmentation rays were used to avoid partial volume effects and successfully remove stool. Clinical results were not reported. Our study did not require dietary restrictions, and contrast material was delivered in convenient capsules.
In summary, a custom colon phantom to evaluate the unprepared colon proved
useful for testing various stool subtraction algorithms. Image processing
tools combining thresholding, expansion, and convolution proved most useful
for stool subtraction. Ninety percent of patients with large colon lesions
( 1 cm in diameter) were identified by an experienced radiologist using
either unsubtracted or stool-subtracted images. Sensitivity and specificity
for smaller lesions, which are especially problematic when stool subtraction
is used, are lower. The role of stool subtraction is equivocal. In view of the
reluctance of patients to accept bowel preparation, further study of this
technique is warranted. Potentially, millions of patients could be spared
colorectal cancer if this technique were improved and widely available.
Acknowledgments
We thank Debora Shreve for her dedication in preparing the manuscript.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
50 years: United States, 2001. Morb
Mortal Wkly Rep 2003;52:193
-196[Medline]
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  A. H. de Vries, H. W. Venema, J. Florie, C. Y. Nio, and J. Stoker Influence of Tagged Fecal Material on Detectability of Colorectal Polyps at CT: Phantom Study Am. J. Roentgenol., October 1, 2008; 191(4): W181 - W189. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 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]
|
|
|
|
|
|
R. J. Stanley The Challenge of Managing Prostate Cancer Am. J. Roentgenol., January 1, 2007; 188(1): 1 - 1. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
