Segmentation of thoracic computed tomography images

Segmentation of thoracic computed tomography (CT) images is an important step in many medical imaging applications. This paper presents an automatic scheme for identifying the patient's contour, the lungs, the trachea and the spinal canal on a set of two-dimensional (2D) thoracic CT images. Three different methods were proposed for the segmentation process. An adaptive thresholding method was used for the delineation of the external skin surface of the patient A 3D credal filter, based on the belief functions theory, was implemented for the lungs and the spinal canal segmentation. Because of the differences between the organs shape, the filter parameters were different. For the lung, no direction was privileged, whereas for the spinal canal the perpendicular direction of the transverse slices was privileged in order to reinforce the inter-slice contribution. A 3D region growing method was used for the trachea segmentation. Segmentation results on a set of 2D CT images are presented and allow highlighting the performances of the proposed methodology. The contours were evaluated by an experimented radiation oncologist.