An automatic colonic polyp detection method based on global normal convergence approaches

Abstract

In this paper, we proposed a new approach for colonic polyp detection based on global normal convergence (GNC) and three dimensional directional shape feature (3D-DSF). The GNC is proposed to compute convexity and convergence of surface as an alternative to traditional curvature based methods. In addition, 3D-DSF extraction has been suggested to extract 3D shapes and morphology feature along specific direction, i.e. from top of the polyp towards its base. This directional feature extraction aids to make unique point of view of polyps regardless of size, orientation and position. The proposed polyp detection is appeared computationally efficient (typically takes 4min per dataset) and it shows 90.9% sensitivity for detection of real polyps larger than 10mm and 86.6% sensitivity for polyps between 6 to 10mm with an average of 7.1 false positives per dataset. The experimental data indicates that the GNC is an effective tool to extract convex region with flexible radius. In addition, 3D directional features extraction are found as an effective set of feature to examine the span of object in specific orientation. The algorithm is being modified and further validated by the clinical experiment. © 2008 Springer-Verlag.