Evaluation of a segmentation procedure to delineate organs for use in construction of a radiation therapy planning atlas.

Objectives: This paper evaluates a semi-automatic segmentation procedure to enhance utilizing atlas based treatment plans. For this application, it is crucial to provide a collection of 'reference' organs, restorable from the atlas so that they closely match those of the current patient. To enable assembling representative organs, we developed a semi-automatic procedure using an active contour method.

Method: The 3D organ volume was identified by defining contours on individual slices. The initial organ contours were matched to patient volume data sets and then superimposed on them. These starting contours were then adjusted and refined to rapidly find the organ outline of the given patient. Performance was evaluated by contouring organs of different size, shape complexity, and proximity to surrounding structures. We used representative organs defined on CT volumes obtained from 12 patients and compared the resulting outlines to those drawn by a radiologist.

Results: A strong correlation was found between the area measures of the delineated liver (r = 0.992), lung (r = 0.996) and spinal cord (r = 0.81), obtained by both segmentation techniques. A paired Student's t-test showed no statistical difference between the two techniques regarding the liver and spinal cord (p > 0.05).

Conclusion: This method could be used to form 'standard' organs, which would form part of a whole body atlas (WBA) database for radiation treatment plans as well as to match atlas organs to new patient data.