Principal Curvature-Based Colonic Polyp Detection Dongqing Chen a , Aly A. Farag a , M. Sabry Hassouna b and Robert Falk c a Department of Electrical & Computer Engineering, University of Louisville, Louisville, KY, 40217, USA b Vital Images, Minnetonka, MN, 55343, USA c Department of Medical Imaging, Jewish Hospital, Louisville, KY, 40202, USA E-mail: farag@cvip.louisville.edu ; URL www.cvip.louisville.edu Abstract. An automatic polyp detection framework is presented based on surface curvatures, and a new color coding scheme is used to highlight the detected polyps which enhances the visualization. An accurate estimate for the surface curvature together with three geometric features is used to detect the polyps. We place the detected polyps inside a polygonal dataset with exactly the same topology and geometry properties as the mesh surface of real colon dataset, and assign different colors to the two separated datasets, in order to highlight the polyps. This approach is shown to detect polyps of size above 5 mm and reduces false positives. The algorithm is validated on simulated data sets as well as real CT colonoscopy data. Keywords: computer aid diagnosis, geometric feature, principal curvatures, colonic polyp detection, and color coding 1. Introduction Colorectal cancer is the second leading cause of cancer-related death and the third most common form of cancer in the United States [1]. Since colorectal cancer is largely preventable, early detection and removal of colonic polyps can substantially reduce the risk of cancer occurrence. Computer tomographic colonoscopy (CTC) is a rapidly evolving diagnostic tool for the location, detection and identification of benign polyps on the colon wall in the early stage before their malignant transformation. As shown in Figure 1, colonic polyps generally grow from the colonic mucosa, and appear as dome-like structure with small curvature. The haustral folds appear like shape ridges, which have large curvature values. The colon walls have nearly flat or cup structures with small curvatures. Different geometric feature descriptors have been used for colonic polyp detection [2-5]. In CTC examination, image display formats can affect the performance of the radiologists. Index color coding is one of the important methods to improve the visualization effect of data. However, the lookup table is sensitive to the hue, saturation, value, and alpha opacity (HSVA). Since it is not easy to know how the lookup table works and how to balance HSVA values, it is hard to understand all the associated parameters. In this paper, we propose a general framework for colonic polyp detection using three geometric features: shape index, curvedness and sphericity ratio. We also include our newly developed color coding approach to enhance the visualization [6]. This paper extends that concept and tests the overall framework on real dataset. 2. Curvature Based Polyp Detection Let p = (x, y, z) denote a voxel on the 3D surface S, if we know the Gaussian curvature and mean curvature, two principal curvatures can be computed as: K H H − + = 2 1 κ and K H H − − = 2 2 κ (1) The main difficulties associated with computing the principal curvatures are the following: 1) under discrete cases, H 2 -K in Eq. 1 is not always guaranteed to be greater