Prior-knowledge treatment planning for volumetric arc therapy using feature-based database mining

Treatment planning for volumetric arc therapy (VMAT) is a lengthy process that requires many rounds of optimizations to obtain the best treatment settings and optimization constraints for a given patient's geometry. We propose a feature-selection search engine that explores previously treated cases of similar anatomy, returning the optimal plan configurations and attainable DVH constraints. Using an institutional database of 83 previously treated cases of prostate carcinoma treated with volumetric-modulated arc therapy, the search procedure first finds the optimal isocenter position with an optimization procedure, then ranks the anatomical similarity as the mean distance between targets. For the best matching plan, the planning information is reformatted to the DICOM format and imported into the treatment planning system to suggest isocenter, arc directions, MLC patterns, and optimization constraints that can be used as starting points in the optimization process. The approach was tested to create prospective treatment plans based on anatomical features that match previously treated cases from the institution database. By starting from a near-optimal solution and using previous optimization constraints, the best matching test only required simple optimization steps to further decrease target inhomogeneity, ultimately reducing time spend by the therapist in planning arcs' directions and lengths.