Ultrasonographic Segmentation of Cervical Lymph Nodes Based on the Graph Cut with Elliptical Shape Constraint / 航天医学与医学工程

ABSTRACT

Objective To propose a modified graph cut algorithm under the elliptical shape constraint to segment the ultrasonogram of cervical lymph nodes. Methods Since cervical lymph nodes were usually oval-shaped, the prior of the elliptical shape information, which was expressed as a distance function, was used to constrain the cut cost of the graph cut algorithm. The initially segmented contour was fitted with an ellipse to get the constraint. Then a certain segmentation was obtained by minimizing the cut cost function constrained with the elliptical prior. This procedure was iterated until it to be convergent. Results Under the same user input, this algorithm successfully segmented nodes on the low contrast ultrasonograms, while the traditional graph cut approach failed. For 20 ultrasonograms, with the average minimum Euclidean distance (AMINDIST) measure and the Hausdorff distance measure, these two distances between contours manually delineated by the radiologist and those segmented by this algorithm were far less than the distances resulted from the traditional graph cut algorithm. Conclusion It is indicated that utilizing the elliptical shape constraint prior can obviously improve the graph cut algorithm for segmenting cervical nodes on ultrasonograms.