Automatic segmentation of abdominal blood vessels from contrasted X-ray CT images

Segmentation of vessel structures in 3D volume data is of great interest for diagnosis and surgical planning. There are a number of methods that employ various intensity-based, textural, or geometric features for vessel extraction from 3D volume data. However, these methods are not successful in the low-contrast and inhomogeneous environments, especially in case of thinner blood vessels. In this study, we propose a simple procedure which automatically extracts vessel structures from abdominal CT images. Initial vessel and bone image is obtained by using thresholds on the preprocessed CT image. First threshold is computed by discriminant analysis of the reduced CT volume, obtained empirically. Second threshold is determined finding the first local minimum in the histogram of reduced data. Then larger vessels (aorta and nearby) are segmented by 3D region growing of the preprocessed CT volume. This image is subtracted from the initial bone and vessel image to obtain a new binary image without larger vessels. Observation of the slice of the resultant labeled image confirms that components associated with bones have a number of holes, while the same for vessels do not. We therefore propose a new structural measure, called “Hole Area Index (HAI)”, which is computed slice-by-slice on the labeled image and is used to isolate bone and soft-tissue regions. The vessel binary is, finally, combined with the pre-computed larger vessels to segment abdominal vessels. A preliminary experiment with ten CT cases shows promising results of segmenting abdominal vessels.