A supervised method is presented for the detection and segmentation of ribs in computed tomography (CT) data. In a first stage primitives are extracted that represent parts of the centerlines of elongated structures. Each primitive is characterized by a number of features computed from local image structure. For a number of training cases, the primitives are labeled by a human observer into two classes (rib vs. non-rib). This data is used to train a classifier. Now, primitives obtained from any image can be labeled automatically. In a final stage the primitives classified as ribs are used to initialize a seeded region growing process to obtain the complete rib cage.
Automatic rib segmentation in CT data
J. Staal, B. van Ginneken and M. Viergever
Computer Vision Approaches to Medical Image Analysis and Mathematical Methods in Biomedical Image Analysis 2004;3117:193-204.