HEAD AND NECK CANCER PATIENT SIMILARITY BASED ON ANATOMICAL STRUCTURAL GEOMETRY

As we develop radiation treatment planning systems for head and neck cancer patients, there is a need to identify reference patients whose anatomical structures share similar features. By finding previously treated patients or prototypical models with the most similar anatomy in the head and neck region, our prototype system can do a better job of identifying lymph node regions based on known regions predefined for similar patients. This can potentially expedite the radiation treatment planning process which currently depends on oncologists manually delineating lymph node regions for each patient slice by slice on their CT scans. Our methodology uses the identification of landmark anatomical structures from the CT scan of the cancer patient and the computation of a Hausdorff distance between these and the corresponding structures of the reference scans. The combination of these structural correspondence features and a set of global shape features provide a distance metric that can rapidly find the most similar reference patient case and thus efficiently and accurately generate the desired lymph node regions for a given test case.