Automatic Detection of Air Holes Inside The Esophagus in CT Images

Abstract

Automatic esophagus segmentation is a challenging problem. The wall of the esophagus consists of muscle tissue, which has a low contrast to vessels, other muscles and lymph nodes. Shape and appearance can vary a lot. It appears solid if it is empty, but it can also be filled with air, remains of orally given contrast agent, or both. Even for a human, it is often impossible to accurately delineate the boundaries given only a single slice. Up to now, the amount of publications on the topic is limited. It combines a spatial prior of the esophagus centerline with a histogram based appearance model. The centerline is extracted using a shortest path algorithm. Then, ellipses are fitted into axial slices by optimizing an energy function that is again histogram based and also has a regularization term for smooth transitions between neighboring slices. The method is semiautomatic and requires two manually placed points on the centerline and also a segmentation of the left atrium and the aorta as input. In another semiautomatic segmentation method is proposed which also uses a spatial prior of the esophagus centerline. The prior is estimated relative to a set of axial 2-D contours of vertebrae, the trachea, the left main bronchi, the aorta and the heart that were segmented manually in seven reference slices. This is combined with a level set segmentation, which is initialized with the detected centerline. In, contour lines that were manually drawn in axial slices are interpolated in the frequency domain without using the image itself. In the user draws one contour in an axial slice, and registration based on optical flow is used to propagate the contour to neighboring slices.