Model-based Segmentation of Pathological Lymph Nodes in CT Data

For the computer-aided diagnosis of tumor diseases knowledge about the position, size and type of the lymph nodes is needed to compute the tumor classification (TNM). For the computer-aided planning of subsequent surgeries like the Neck Dissection spatial information about the lymph nodes is also important. Thus, an efficient and exact segmentation method for lymph nodes in CT data is necessary, especially pathological altered lymph nodes play an important role here. Based on prior work, in this paper we present a noticeably enhanced model-based segmentation method for lymph nodes in CT data, which now can be used also for enlarged and mostly well separated necrotic lymph nodes. Furthermore, the kind of pathological variation can be determined automatically during segmentation, which is important for the automatic TNM classification. Our technique was tested on 21 lymph nodes from 5 CT datasets, among several enlarged and necrotic ones. The results lie in the range of the inter-personal variance of human experts and improve the results of former work again. Bigger problems were only noticed for pathological lymph nodes with vague boundaries due to infiltrated neighbor tissue.