Automatic pancreas segmentation in contrast enhanced CT data using learned spatial anatomy and texture descriptors

Pancreas segmentation in 3-D computed tomography (CT) data is of high clinical relevance, but extremely difficult since the pancreas is often not visibly distinguishable from the small bowel. So far no automated approach using only single phase contrast enhancement exist. In this work, a novel fully automated algorithm to extract the pancreas from such CT images is proposed. Discriminative learning is used to build a pancreas tissue classifier that incorporates spatial relationships between the pancreas and surrounding organs and vessels. Furthermore, discrete cosine and wavelet transforms are used to build computationally inexpensive but meaningful texture features in order to describe local tissue appearance. Classification is then used to guide a constrained statistical shape model to fit the data. Cross-validation on 40 CT datasets yielded an average surface distance of 1.7 mm compared to ground truth which shows that automatic pancreas segmentation from single phase contrast enhanced CT is feasible. The method even outperforms automatic solutions using multiple-phase CT both in accuracy and computation time.