Data driven editing of RIB centerlines

Abstract

Rib Unfolding refers to the flattening of the rib cage into a two dimensional image that enables the radiologist to quickly examine all ribs for metastases and fractures without having to scroll through every single image in the CT scans. The rib centerline extraction is the core component of the rib unfolding technology. If the computed centerlines deviate from the true centerlines of the ribs then the unfolding shows the bone cortex or even surrounding tissues, making it impossible to spot lesions inside the ribs. Therefore, this paper presents an interactive system for data driven editing of rib centerlines. The user simply has to place a click point at the center of a rib. From there, the centerline is automatically corrected in both directions until the distance to the old one is below a threshold. We formulate the interactive data driven editing as an energy minimization problem where the new centerline point is calculated as the center of mass of the segmentation mask in the plane orthogonal to the rib centerline, this segmentation mask represent the rib cross section. An automatic tracing scheme is calculated based on the refined centerline point and along the tangent vector to the refined centerlines to get the next orthogonal plane to be segmented. The novelty of this paper is three fold: 1. It defines an intuitive minimal user interaction workflow for the correction of rib centerlines. 2. It uses the data in the neighborhood of the centerline point to formulate the energy minimization problem and obtain a more accurate segmentation. 3. We present a novel re-initialization component to prevent the rib tracing from deviating from the correct solution. The initial point to be corrected is re-initialized based on the neighboring ribs. Quantitative assessment of our method shows that in 85% percent of the cases the rib correction can be perfomed using one or two points.