{"id":4058,"date":"2011-02-24T11:43:17","date_gmt":"2011-02-24T16:43:17","guid":{"rendered":"https:\/\/www.med.unc.edu\/radonc\/?p=4058"},"modified":"2020-04-29T15:30:13","modified_gmt":"2020-04-29T19:30:13","slug":"automatic-segmentation-of-the-male-pelvis-for-igrt-and-art","status":"publish","type":"post","link":"https:\/\/www.med.unc.edu\/radonc\/automatic-segmentation-of-the-male-pelvis-for-igrt-and-art\/","title":{"rendered":"Automatic segmentation of the male pelvis for IGRT and ART"},"content":{"rendered":"

Goal: to reduce user variability that can affect clinical decisions by statistically training shape models representing internal organs for better targeting.<\/p>\n

\n

\"COLabCritical treatment planning and delivery decisions in radiation therapy include aiming and shaping radiation beams to hit the tumor target and spare nearby normal organs. CT images provide a geometric model of the patient that is used to make these decisions with the aid of computer-based planning systems, sometimes on a daily basis over several weeks. Current methods to extract the patient model from CT images in clinical practice require expert human interaction and thus are extremely time consuming and expensive. Moreover humans, even experts with similar training, are known to exhibit inter- and intra-user variabilities that can adversely affect clinical decisions. Automatic methods are being developed based on a class of so-called statistically trainable deformable-shape models (SDSMs) called m-reps (m<\/u>edial rep<\/u>resentations). An m-rep is an explicit mathematical representation (a model) of an anatomical object that can be automatically imbedded in a target image, rigidly registered with the corresponding target object, and then deformed to closely match the shape of the specific target object. To accomplish all this, m-reps must be trained against models built by expert humans to generate statistical probability distributions that define the range of object shapes (e.g., prostates, bladders and rectums) and the range of each object\u2019s intensity patterns in medical images across the human population of interest.<\/p>\n

M-reps have special capabilities for representing shape and for training probability distributions that cannot be matched by potentially competing SDSMs. Pre-clinical research at UNC shows that m-reps rival expert humans in extracting anatomical geometry from medical images. External collaborators include Memorial-Sloan Kettering Cancer Center and William Beaumont Hospital.<\/p>\n

Selected Publications<\/h4>\n