Thorax Autosegmentation for Radiotherapy of Lung Cancer

DESCRIPTION (provided by applicant): SBIR Phase 1 Application entitled Thorax Auto segmentation for Radiotherapy of Lung Cancer As attested by the attached letters from Morphormics partners manufacturing radiotherapy equipment, there is a need for a commercial product that significantly outperforms presently available products for auto segmentation of thorax objects at risk in radiotherapy of lung cancer. Morphormics develops superior auto segmentation software, with a focus on radiotherapy. Our methods learn and use anatomic (geometric) knowledge via the novel m-reps representation, and they learn and use image appearance information via our novel methods based on quantile functions. Based on our successes in the auto segmentation of male pelvis objects from CT for planning the treatment of prostate cancer, we plan to build a product for segmenting all of the objects at risk (OARs) for lung cancer radiotherapy via these techniques. For this Phase 1 period, we have chosen to demonstrate our abilities with three of these OARs such that the success for these will be persuasive that we can achieve success for the others in a Phase 2 proposal. In particular, in the proposed Phase 1 work we will develop and evaluate methods of automatic segmentation of the lung, upper bronchial tree, and heart requiring less than 1 minute of user interaction time and 1-2 minutes of computation time per structure to produce clinically usable results, as demonstrated in more than 100 test cases. PUBLIC HEALTH RELEVANCE: Relevance to Image-Guided Cancer Interventions of SBIR application entitled Thorax Auto segmentation for Radiotherapy for Lung Cancer Lung cancer is a leading cause of death, and radiotherapy is a common form of treatment. Planning radiotherapy of lung tumors requires the segmentation (extraction of the location) from planning CT images of many thoracic objects at risk (OARs) for radiation damage. Knowing where these objects are allows prescribing radiation that is focused on the tumor and away from these OARs. The present common method for segmentation, done manually slice by image slice, is inherently time-consuming and sometimes inaccurate, given clinical time constraints. Auto segmentation is an important tool to provide at least as accurate, more robust and reproducible, and faster segmentation. However, there is no commercial software for auto segmentation of these thorax OARs that combines accuracy and convenience in such a way as to be clinically competitive with manual segmentation. It is the objective of this proposal to begin building a product that accomplishes this aim. Our company focuses on the development of products for auto segmentation from medical images, with a focus on radiotherapy. Our novel methods have led to a product that achieves unique success in auto segmenting structures in the male pelvis from CT in the treatment of prostate cancer with clinically required accuracy, robustness, and speed. This product has been approved by the FDA as part of the radiation treatment planning system of a major manufacturer of radiotherapy equipment, and we anticipate that it will be helping the treatment of real patients in early 2010. The success of our methods for male pelvis segmentation, together with the similarity of many of the thorax structures and image contrast challenges in the thorax, leads to the expectation that similar success can be achieved to produce clinically useful auto segmentations of structures in the thorax from CT. A commercial product that can provide these auto segmentations will lead to improved therapy of lung cancer. However, seed funding by an SBIR grant is necessary to get this thorax segmentation product started.