Automatic Rib Segmentation in Chest CT Volume Data

An automatic segmentation method for extraction of human rib structures from chest CT volume data is presented. Segmentation is initiated from the middle coronal slice to attain complete and isolated 12 pairs of ribs with a recursive tracking on coronal slices spreading from the middle coronal slice. At each coronal slice, the lung contours are extracted, candidate rib regions are derived from thresholding, and refined by adding constraints on shape and location with respect to the lung contour and centroids of rib regions of the former coronal slice. Anatomical and radiological prior knowledge has been explored to ignore those rib regions connected to spine for breaking the connection with spine. Appropriate thresholds are chosen so that lung regions can be binarized, the cartilage connecting ribs and sternum are binarized as background to break the connection between the sternum and ribs. Our method is tested on 15 CT data sets. Experiments show that radiologists are satisfied with the extracted rib regions on coronal slices, and only those rib regions connected to spine are discarded (5.5% of all 2D rib regions). The method provides a foundation for further investigation on computer-aided diagnosis of rib fractures.