Automatic Segmentation of the Human Abdomen on Clinically Acquired CT

Abstract

The human abdomen is an essential, yet complex body space clinically. Computational tomography (CT) scans are routinely taken for the diagnosis and prognosis of abdomen-related diseases, such as the pathological injuries or changes of abdominal organs, and the abnormal extrusion through the abdominal wall. Segmentation on CT images provides a computational representation for the structures of interest to access the structural characteristics, and thus establishes a foundation for quantitative analysis. While fully automated segmentation on large-scale clinical imaging data has been the target of intense efforts for decades, robust segmentation systems for the abdomen remain elusive. Here, we present automatic segmentation approaches for (1) the abdominal wall (covering both outer and inner surfaces over the range between xiphoid process and pubic symphysis) and (2) multiple abdominal organs (up to 13 organs, including liver, spleen, and kidneys) on clinically acquired CT. State-of-the-art atlas- and surface-based image processing techniques are investigated and robustly adapted to the challenging problems in abdomen given (a) anatomical structures with substantial occurrences of abnormalities and large variations in shape and appearance, and (b) CT scans with varied sizes and resolutions, fields of view, contrast enhancement, and imaging artifacts. Translational studies are performed to demonstrate the efficacy of the presented segmentation to support clinical decisions.