Geometric and nomenclature errors are commonly encountered in automated treatment planning. We describe a novel algorithm to extract organ geometry relationships from patient structure DICOM data to construct a database that can be used to detect organ contour inaccuracies including relational and naming errors. Twenty-five sets of head and neck patients' treatment plan data (CT, structures) were retrospectively retrieved from our institution. For each dataset, various organs were contoured and verified by experienced physicians. The relative position and orientation between organs were extracted from each patient and the data were used to construct an organ relationship database model. The model was tested using a dataset originating from an in-house organ renaming software that often-introduced organ contour naming mismatches. As part of the validation test, the renamed organs relative positions were compared with the database model to identify mismatches. Within the forty head and neck patients, we extracted the geometric relationship between 201 organ pairs. The average number of unique types of organ pairs (for example, left parotid with left eye is one type of organ pair) stored in the database was 12. Fifteen head and neck structure sets automatically renamed using our in-house organ renaming tool was used as validation data. All of the 30 random assigned wrong name labels present in these structure sets were identified using the established organ geometry relationship database. We successfully constructed a head and neck organ geometry relationship database and validated it in a contour naming quality assurance process. This novel scheme can be expanded to the entire body and shows a great potential in automatic plan physics QA procedure. It should be one essential QA step in an autonomous treatment planning process.
Keywords: Automatic contour QA; Geometry relationship; Mathematical model.
Copyright © 2022 American Association of Medical Dosimetrists. Published by Elsevier Inc. All rights reserved.