International Journal of Software Engineering and Its Applications Vol. 3, No. 1, January, 2009
Automatic Colonic Polyp Detection by the Mapping using Regional Unit Sphere 1
Park M., 1Jin S.J., 2Hofstetter R., 1Xu M., 3Kang B.H. 1. School of Design, Communication and IT, The University of Newcastle, Australia 2. IntelliRad Solution, East Hawthorn 3122, Victoria, Australia 3. School of Computing, The University of Tasmania, Australia
[email protected] Abstract Colonic polyps appear like elliptical protrusions on the inner wall of the colon. The many proposed algorithms assumed the shape of a polyp as a spherical cap, so the algorithms are not flexible when the polyps are irregular shapes. In this paper, we propose a mapping using regional unit sphere (MuRUS) method to overcome the problem caused by unexpected polyp shapes. The MuRUS has shape invariant and size invariant properties. Our method was applied to colon CT images from 37 patients each having a prone and supine scan. There are 45 colonscopically confirmed polyps. The results obtained by our algorithm were compared with those gold standards. 100% of polyps >= 10mm in diameter were detected, 90% of polyps >= 6mm in diameter were detected and 70% of polyps < 6mm in diameter were detected at 7.0 FPs per patient.
1. Introduction Colorectal cancer is the most commonly diagnosed, non-cutaneous cancer in Australia. In 2001 there were 12,844 cases of colorectal cancer (6,961 in men and 5,883 in women) and 4,754 deaths (2,601 in men and 2,153 in women). In Australia, the lifetime risk of developing colorectal cancer before the age of 75 years is approximately one in 17 for men and one in 26 for women [1]. According to the American Cancer Society [2, 16, 17], the odds of a cure are higher than 90 percent when tumors in the colon are found and treated early. Cure rates are highest when a tumor is localized, meaning that the cancer has not yet spread to other parts of the body. One of the biggest problems with colon cancer is that most cancers produce no symptoms early on when they are small and limited to the colon. Therefore, colon cancer is an ideal disease for screening. Screening is the process of looking for a cancer in healthy persons who are at risk for the disease but have no symptoms. Colonoscopy can allow a doctor to see and find 100% of polyps and potential cancers. However, screening for colon cancer lags far behind screening for other cancers, since colonoscopy is expensive, invasive, uncomfortable and carries a risk of complications, including bleeding and perforation of the colon. Computerized tomography (CT) colonography, commonly called virtual colonoscopy, has shown great promise as a non-invasive visualization technique to search for potentially cancerous polyps since there is no risk of bleeding or colon perforation and intravenous sedation is unnecessary [3]. Physicians hope it will encourage more people to be screened for colon cancer. However, the accuracy and efficiency of viewing hundreds of CT images per exam are limited by human factors such as attention span and eye fatigue. The review of hundreds of images for each patient is tiring and time-consuming. That is where computer-aided detection (CAD) comes into the picture[18].
11
International Journal of Software Engineering and Its Applications Vol. 3, No. 1, January, 2009
1a
1b
1c
2a
2c
1d
2b
2d
Figure 1. 1a. A polyp in WRAMC VC-386, 1b. the segmented polyp (1a) by the CoLN method, 1c. 3D illustration of a results by the MuRUS method, 1d. 2D projection illustration of 1c. 2a. A polyp in WRAMC VC-241, 2b. the segmented polyp (2a) by the CoLN method, 2c. 3D illustration of a results by the MuRUS method, 2d. 2D projection illustration of 2c. The red numbers, i, 1= 6mm in diameter were detected and 70% of polyps < 6mm in diameter were detected at 7.0 FPs per patient.
6. Discussions The main key of the MuRUS is the regoning. There are many ways to mathematically divide the surface of a sphere. However, the MuRUS is based on the orientations, (φ,θ), so the sphere should be divided based on the angles from the origin. Another issue is the size of region. To represent the information from the normals of the candidate surface, the number of regions and the size of region should be set carefully. If there are too many regions or too less regions, the state sequence obtained from the regions does not represent the object. We empirically select 26 regions. These are one around the North Pole, one around the South Pole, and between them we divide into 3 layers. The middle 3 layers, which parallel to the xy plan, could be divided into 4 directions such as (positive x, positive y), (positive x, negative y), (negative x, positive y), and (negative x, negative y). However, this dividing is only useful for cubical shapes. Therefore, we divide each above direction into 2 directions, so the regions capture the information of flat or curve shape. Because the MuRUS is constructed solely from normals, the triangluarisation algorithm affects the accurateness of the MuRUS. The HMM is often used in finding patterns which appear over a space of time, but we uses the HMM to find the patterns which appear over the regional unit sphere, since the MuRUS has the Markov property. We adopt the Hidden Markov Model (HMM) to utilize the Markov property of our MuRUS method which is the probability of a given state coming up next, pr(xt=Si), and this may depend on the prior history to t-1. This property allows our algorithm to detect any irregular cap shape and various sizes of objects.
7. References [1] A guide for general practitioners, (2006) Clinical Practice Guidelines for the Prevention, Early Detection and Management of Colorectal Cancer, http://www.nhmrc.gov.au/publications/synopses/cp106/cp106divided.htm [2] ACS: http://www.cancer.org
16
International Journal of Software Engineering and Its Applications Vol. 3, No. 1, January, 2009
[3] RSNA 2005 News (2005): Virtual Colonoscopy Performance Enhanced by Computer Aided Detection, http://www.rsna.org/rsna/media/pr2005/virtual_colonoscopy.cfm [4] Vining D.J., Ge Y., Ahn D.K. and Stelts D.R. (1999): Virtual Colonoscopy with Computer-Assisted Polyp Detection. Computer-Aided Diagnosis in Medical Imaging, 445-452 [5] Summers R.M, Beaulieu C.F., Pusanik L.M., Malley J.D., Jeffrey Jr. R.B., Glazer D.I. and Napel S. (2000): Automated polyp detector for CT colonography: Feasibility study, Radiology, 216:284-290 [6] Summers R.M., Johnson C.D., Pusanik L.M., Malley J.D., Youssef A.M. and Reed J.E. (2001): Automated polyp detection at CT colonography: Feasibility assessment in a human population, Radiology, 219:1261-1272 [7] Yoshida H. and Nappi J. (2001): Three-dimensional Computer-aided diagnosis Scheme for detection of Colonic Polyps, IEEE Transactions on Medical Imaging, 20:1261-1274 [8] Yoshida H., Masutani Y., MacEneaney P., Rubin D.T. and Dachman A.H. (2002): Computerized Detection of Colonic Polyps at CT Colonography on the Basis of Volumetric Features: Pilot Study, Radilogy, 222:327-336 [9] Paik D.S., Beauliey C.F., Rubin G.D., Acar B., Jeffrey Jr. R.B., Yee J., Dey J. and Napel S. (2004), Surface Normal Overlap: A Computer-Aided Detection Algorithm with Application to Colonic Polyps and Lung Nodules in Helical CT, IEEE Transactions on Medical Imaging, 23(6):661-675 [10] Kiss G., Cleynenbreugel J.V., Thomeer M., Suetens P. and Marchal G. (2002): Computer-Aided Diagnosis in Virtual Colonography via Combination of Surface Normal and Sphere Fitting Methods, European Radiology, 12:77-81 [11]Gokturk S.B., Tomasi B., Acar B., Beaulieu C.F., Paik D.S, Jeffrey Jr. R.B., Yee J. and Napel S. (2001): A Statistical 3-D Pattern Processing Method for Computer-Aided Detection of Polyps in CT Colonography, IEEE Transactions on Medical Imaging. 20:1251-1260 [12] Li J., Huang A., Yao J., Bitter I., Petrick N., Summers M.R., Pickhardt J.P. and Choi R.J. (2006), Automatic Colonic Polyp Detection using Multiobjective Evolutionary Techniques, Medical Imaging: Image Processing, Proc. of SPIE 6144, 61445E [13] Horn BK. (1984) Extended Gaussian Images, Image Understanding Workshop, Defense Advanced Research Projects Agency, Science Appkications International Corp., 72-89 [14] Rabiner LR. (1989) A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition, Proc. IEEE, 77(2)257-285 [15] Park M., Hofstetter R. and Luo S. (2006) Automatic Polyp Detection in CT Colongraphy, Visual Information Processing, 135-141 [16] Petty T.L. (2000): Screening strategies for early detection of lung cancer: The time is nsw, JAMA, 284:19771980 [17] Bulman W. (2004): Screening for Colon Cancer: When Early Detection Can Mean a Cure, http://www.dentalplans.com/Dental-Health-Articles/Screening-for-Colon-Cancer-When-Early-Detection-CanMean-a-Cure.asp [18] RSNA 2001 News (2001): Colon Cancer Screening Improved with Computer as a ‘Second Set of Eyes’, [19] Nappi J. and Yoshida H. (2002): Automated Detection of Polyps with CT Colonography: Evaluation of Volumetric Features for Reduction of False-Positive Findings, Academic Radiology, 9:386-397 [20]Lin Z., Jin J.S. and Talbot H., (2001): Unseeded Region Growing fro 3D image Segmentation, Conferences in Research and Practice in Information Technology, 2:31-38
Authors Mira Park received the M.S. and PhD degree in the computer science engineering from the University of New South Wales in 2003. During 2004~2005, she worked as a research fellow in the computer science and software engineering at the University of Melbourne. She is a research fellow in the school of design, communication and IT of the Newcastle University.
17
International Journal of Software Engineering and Its Applications Vol. 3, No. 1, January, 2009
18