Classification of the Colonic Polyps in CT-Colonography Using Region Covariance as Descriptor Features of Suspicious Regions
- 106 Downloads
We present an algorithm to classify polyps in CT colonography images utilizing covariance matrices as object descriptors. Since these descriptors do not lie on a vector space, they cannot simply be fed to traditional machine learning tools such as support vector machines (SVMs) or artificial neural networks (ANNs). To benefit from the simple yet one of the most powerful nonparametric machine learning approach k-nearest neighbor classifier, it suffices to compute the pairwise distances among the covariance descriptors using a distance metric involving their generalized eigenvalues, which also follows from the Lie group structure of positive definite matrices. This approach is fast and discriminates polyps from non-polyps with high accuracy using only a small size descriptor, which consists of 36 unique features per image region extracted from the suspicious regions that we have obtained by combined cellular neural network (CNN) and template matching detection method. These suspicious regions are, in average, 15 × 17 = 255 pixels in our experiments.
KeywordsCT colonography Colonic polyp detection Covariance descriptor
This research is supported by Istanbul University, Research Fund. Project No: T-502.
- 1.Summers, R. M., Johnson, C., Pusanik, L. M., Malley, J. D., Youssef, A., and Reed, J., Automated polyp detector at CT colonography: feasibility assessment in a human population. Radiology. 219:51–59, 2001.Google Scholar
- 3.Macari, M., Virtual colonoscopy: clinical results. Semin. Ultrasound CT. 22(5):432–442, 2001.Google Scholar
- 8.Kilic, N., Ucan, O. N., and Osman, O., Automatic colon segmentation using cellular neural network for the detection of colorectal polyps. IU-JEEE. 7:419–423, 2007.Google Scholar
- 10.Li, J., Franaszek, M., Petrick, N., Yao, J., Huang, A., Summers, R. M., Wavelet method for CT Colonography computer-aided polyp detection. Proc. IEEE International Symposium on Biomedical Imaging (ISBI): From Nano to Macro, pp:1316–1319, Arlington, VA, USA, 6–9 April 2006.Google Scholar
- 13.Kilic, N., Ucan, O. N., Osman, O., Colonic polyp detection in CT colonography with fuzzy rule based 3D template matching. J. Med. Syst. 2008. doi: 10.1007/s10916-008-9159-3.
- 14.Porikli, F., and Kocak, T., Robust license plate detection using covariance descriptor in a neural network framework. Proc. AVSS. 06:107–112, 2006.Google Scholar
- 15.Forstner, W., Moonen, B., A metric for covariance matrices. Technical report, Dept. of Geodesy and Geoinformatics: Stuttgart University, 1999.Google Scholar
- 16.Tuzel, O., Porikli, F., and Meer, P., Region covariance: a fast descriptor for detection and classification, ECCV 2006. Part II LNCS. 3952:589–600, 2006.Google Scholar
- 18.Jerebko, A. K., Malley, J. D., Franaszek, M., Summers, R. M., Multi-network classification scheme for detection of colonic polyps in CT colonography data sets. Acad. Radiol. 10:154–160, 2003.Google Scholar
- 19.Tuzel, O., Porikli, F., Meer, P., Region covariance: a fast descriptor for detection and classification. In Proc. 9th European Conf. on Computer Vision, Leonardis A., Bischof H., Pinz A (eds.), vol. 2, Lecture Notes in Computer Science 3952, pp. 589–600, Berlin: Springer, 2006.Google Scholar