Lymph node segmentation in CT slices using dynamic programming

An automatically extracted 2d contour of a lymph node in a single CT slice is required for size assessment as well as for the initialization of model based 3d lymph node segmentation algorithms. This paper presents a novel single slice lymph node segmentation approach. It finds a closed path around a seed point thereby minimizing an energy function which depends on gradients, intensities and shape priors. A globally optimal solution is guaranteed to be found using dynamic programming. The approach is evaluated on a set of 49 representative lymph nodes from clinical images for which manual expert segmentation is available. The resulting average dice similarity coefficient of 0.81 is close to manual segmentation (0.89).