Abstract for HONS 05/21
Contour Splitting for Branching Structures in CT Image Reconstructions
Cameron Stevenson
Department of Computer Science and Software Engineering
University of Canterbury
Abstract
Medical imaging modalities such as HRCT and MRI produce stacks of images depicting regions of different tissue. These image stacks can be used to reconstruct the 3D structure of internal organs. The renders or models produced are useful in diagnosis, treatment, education, surgical simulation, robot assisted surgery, and virtual reality. Correspondence methods acting on contours segmented from scan images are one such way of 3D reconstruction. These methods have difficulty in reconstructing structures such as branches and bends. There is ambiguity in contour correspondence, and extra steps must be added to point correspondence. This paper builds upon prior correspondence methods, with a new approach to point correspondence, and specifically aiming to improve on reconstructing these problematic cases. Analysis has been performed on synthetic models and real scan data. The proposed method, "contour splitting and point angle", handles low plane count and branching structures better than prior correspondence methods, giving more accurate reconstructions. On average the proposed method improved in reconstruction accuracy compared to the most related prior method by 15.2%. The main contribution is the new techniques and algorithms developed for point correspondence, which may be useful in future applications.
