Investigating mediastinal lymph node stations segmentation on thoracic CT following experts guidelines

In radiation therapy, accurate delineation of mediastinal lymph node stations on thoracic CT is essential for both prognostication and treatment delivery. We propose an original approach based purely on geometrical considerations, without using grey levels, that follow the reference guidelines and attempt to replicate the delineation undertaken manually by the experts. The proposed method is a greedy process based on fuzzy relative position constraints. It progressively refines an initial region towards the target by using a set of predefined anatomical structures. Experiments were conducted with two CT images that were manually segmented by experts. Average Dice Similarly Coefficient between segmented and references stations was close to 77%. This fast method (30 sec) could potentially assist the expert, for example in detecting situations where the guidelines are not strictly followed. To our knowledge, this is the first time such an approach has been proposed for this problem.