3-D Segmentation of Human Sternum in Lung MDCT Images

A fully automatic novel algorithm is presented for accurate 3-D segmentation of the human sternum in lung multi detector computed tomography (MDCT) images. The segmentation result is refined by employing active contours to remove calcified costal cartilage that is attached to the sternum. For each dataset, costal notches (sternocostal joints) are localized in 3-D by using a sternum mask and positions of the costal notches on it as reference. The proposed algorithm for sternum segmentation was tested on 16 complete lung MDCT datasets and comparison of the segmentation results to the reference delineation provided by a radiologist, shows high sensitivity (92.49 %) and specificity (99.51 %) and small mean distance (dmean=1.07 mm). Total average of the Euclidean distance error for costal notches positioning in 3-D is 4.2 mm.