Colon cancer is the second deadliest cancer in the US.[1,2] Most colon cancers begin as benign polyps that can be found by colonoscopy. Optical colonoscopy is considered the gold standard for finding polyps, and involves passing an endoscope through the rectum and screening the colon. The incident of colon cancer can be reduced by polypectomy using colonoscopy screening.[3] However, optical colonoscopy requires sedation, and is an invasive monitoring technique.

CT colonography (CTC) is a rapidly-evolving, non-invasive substitute for colonoscopy.[3,4,5] The American Cancer Society (ACS) has most recently added CTC study to its screening guidelines for colorectal cancer. Recently, several large-scale, multi-center clinical trials for virtual colonoscopy (ACRIN, IMPACT, etc.) have successfully finished, demonstrating the sensitivity of CTC in the detection of polyps on a par with that of optical colonoscopy.

Electronic colon cleansing (ECC) aims to remove the contrast agent from the CT abdominal images so that a virtual model of the colon can be constructed. Previous work on ECC for prepared colon with liquid tagging is extensive, including: a probabilistic method with special regularization through Markov random field is utilized in [4,5]; fuzzy clustering and deformable models are utilized in [6,7]; and a low level classification based on statistical analysis of intensity values of 3D neighborhood of each voxel is used in [8]; followed by a high level processing technique to extract colon lumen.

However, like optical colonoscopy, these methods require bowel preparation. Fecal tagging of the unprepared colon has been proposed as a method to reduce the discomfort, side effects, and sleep disturbance of bowel preparation.[1] Recent study has shown that ECC can increase the sensitivity and specifity of the CT colonography in the unprepared colon.[9] Previous research on ECC for the unprepared colon is very limited. Carston, et al. proposes an iterative algorithm for stool subtraction, which successfully eliminates untagged parts of the stool.[3] However, it also removes colon folds in the ECC process, imparing the visualization and polyp detection.

We present a new framework for ECC of the unprepared colon. Our framework has three major steps: (1) removing untagged parts of the stool, (2) removing the air-stool partial volume, and (3) reconstructing tissue-stool boundary considering the pseudo-enhancement artifact.

(1) Removing untagged parts of the stool

A binary volume B1, the same size as the CT volume, is generated and all voxel values are initialized to zero. Voxels with intensity more than a threshold T1 are set to one. Threshold T1 is chosen such that all the voxels with intensity of one correspond to stool. A 3D flood fill operation is performed to remove the untagged parts of the stool. As a result, the iselands of low intensity voxels isolated inside the stool are removed, resulting in the binary volume B2.

(2) Removing the air-stool partial volume

The air-stool boundary is identified as follows:
(a) The perimeter of the stool is found by morphological operations on B2
(b) The section of the perimeter touching air is found by casting 18 rays in 3D and looking for air and stool voxels.

The air-stool boundary, which contains the partial volume artifact, is then easily removed

(3) Reconstructing tissue-stool boundary considering the pseudo-enhancement artifact

The colon is filled with pressured air during CT colonography. The deformation of the colon wall due to the air pressure depends on the pressure of the air. Since this pressure is constant across the colon, a prior on the shape of the colon wall can be obtained. Assuming that the colon wall is homogeneous, we obtain a prior for the shape of the colon wall. We use the prior shape to correct the pseudo-enhancement caused by the tagged material.

We have tested the first and second steps of our method using patient data. No laxative drug was administrated to the patients. As a result their colon contained considerable solid stool. The first step successfully removed untagged parts of the stool, which were prevalent because of high level of stool heterogeneity. Partial volume effect at the air-stool boundary was completely removed after the second step. We have proven that our shape, prior due to the air pressure in colon holds with good approximation in CT data obtained from patients who were given laxative drugs. We are currently implementing the algorithm to use this shape prior to correct for pseudo-enhancement effect in CT data acquired with no bowel preparation.

The three contributions of this paper limit the artifacts resulting from the ECC of an unprepared colon. The first step removes untagged parts of stool, which are inevitable because of stool inhompgenity. The flood fill approach eleminates the need for dilation, as is proposed by [3], which removes the colon folds. In the second step, we identify the air-stool partial volume effect and remove it, again without dilation, which can destroy colon folds. In the third step, we correct for pseudo-enhancement without a need for phantom-based calibration, as is required by [12]. Such calibration can be cumbersome, especially since it varies for different CT machine models.

Our three-step solid stool detagging approach successfully removes tagged material with minimal artifacts, opening the avenue for CT colonography with no bowel preparation.

[1] Lefere P, Gryspeerdt S, Dewyspelaere J, Baekelandt M, Van Holsbeeck B. Dietary Fecal Tagging as a Cleansing Method before CT Colonography: Initial Results—Polyp Detection and Patient Acceptance. Radiology. 2002;224:393.
[2] Johnson D, Manduca A, Fletcher J, et al. Noncathartic CT Colonography with Stool Tagging: Performance With and Without Electronic Stool Subtraction. AJR. 2008;190:361-366.
[3] Carston M, Wentz R, Manduca A, Johnson D. CT colonography of the unprepared colon: an evaluation of electronic stool subtraction. SPIE Medical Imaging. 2005;5746:424-431.
[4] Sidney J, Winawer M. Screening of colorectal cancer. Surg Oncol. 2005;14:699-722.
[5] Winawer S, et al. Prevention of colorectal cancer by polypectomy. The National Polyp Study Workgroup. N. Eng. J Med. 1993;329:1977-1981.
[6] Pickhardt P, Choi JHR. Electronic Cleansing and Stool Tagging in CT Colonography: Advantages and Pitfalls with Primary Three-Dimensional Evaluation. AJR. 2003;181:799-805.
[7] Li L, et al. An Image Segmentation Approach to Extract Colon Lumen through Colonic Material Tagging and Hidden Markov Random Field model for virtual colonoscopy. SPIE. 2002;4683:406-411.
[8] Wang Z, et al. An Improved Electronic Colon Cleansing Method for Detection of Colonic Polyps by Virtual Colonoscopy. IEEE Trans. Biomed. Eng. 2006;53:1635-1646.
[9] Yao J, et al. Colonic polyp segmentation in CT colonography-based on fuzzy clustering and deformable models. IEEE Trans. Med. Imag. 2004;23:1344-1352.
[10] Yao J, Summers RM. Adaptive deformable model for colonic polyp segmentation and measurement on CT colonography. Med. Phys. 2007;34:1655-1664.
[11] Cehn, et al. A novel approach to extract colon lumen from CT images for virtual colonoscopy. IEEE Trans. Med. Imag. 2000; 19:1220-1226
[12] Nappi J, Yoshida H. Adaptive correction of the pseudo-enhancement of CT attenuation for fecal-tagging CT colonography. Medical Image Analysis. 2008;413-426.