The " TRACE " Method for Segmentation of Lungs from Chest CT Images by Deterministic Edge Linking

Accurate and robust segmentation of anatomical structures is a basic requirement of image-domain medical expert systems. In particular, when image analysis is based on a model representation of structure, a high level of relational accuracy is required as the diagnosis of a disease process may be based upon a subtle disparity with normal anatomy. This report introduces an implementation for the segmentation of lung tissue from chest CT scans. TRACE represents an object using its edge description. We argue that the explicitly controllable and non-approximating nature of TRACE has some advantages over heuristic techniques such as Active Contour Models (Snakes) for segmentation of non-occluded objects with high complexity edges.