Celiac trunk segmentation incorporating with additional contour constraint

Abstract

The estimation of the vascular direction of celiac trunk plays an important role in the resection of gastric cancer as it can help doctors to plan the specific gastrectomy. The traditional manual estimation is the current clinical standard but time-consuming and subjective. Still, it is a challenging task for current computer vision methods to fully use the volume data in computer-aided clinical diagnosis because the segmented contour of object is of low quality and the small target can’t be completely segmented in the existing methods of convolutional neural network. In this paper, we proposed a novel contour constraint multi-branches network which used an additional contour segmentation branch as constraint for celiac trunk segmentation. The experiments show that our method outperforms other compared methods both quantitively and qualitatively, which provides effective framework for automated, accurate, and reliable vascular direction of celiac trunk estimation to help doctors in clinical diagnosis.