AbstractID: 9152 Title: A Robust 3D Morphological Algorithm for Automatic Skin Contour Detection on CT Images 915222895.doc External skin contouring is the required initial step in many 3D image-based applications, such as treatment planning systems, 3D surface-based fusion/registration applications, and surface rendering algorithms. Local or optimal edge detection algorithms can be used for this purpose, but the post processing steps such as the joint/branch disconnections and the edge classifications are known to be very difficult tasks and these algorithms generally produce open contours. In this study we propose a new automatic method to segment the patient external skin contours on the radiation therapy planning CT images by employing a series of 2D/3D morphological image processing algorithms. The method starts with the iterative global thresholding to binarize the input CT image and the result binary image goes through several morphological algorithms; Opening, Connected Component Extraction, and Boundary Extraction. The Opening disconnects the peripheral objects such as the couch and the head masks and the Connected Component Extraction collects connected pixels and label them. Finally the boundary information is represented using the Polygonal Contour Approximation method. In some cases, it generates artifacts when the couch or the supporting devices share a wide contact with the patient surface. These artifacts are removed by detecting the couch position using Hough Transform. This new method was extensively tested on the CT images of over 500 patients treated in our institute in 2002. The result showed 98% correct segmentations and 25% of them needed the Hough Transform to remove the artifacts. Jinkoo Kim Page 1 3/5/03