A Version-Controlled Workflow for Large-Cohort Creation and Evaluation of Automated Radiotherapy Planning Pipelines
Abstract
Purpose
Automated radiotherapy treatment planning pipelines can reduce planning time and improve standardization, but clinical adoption is often limited by the time and specialized automation expertise required to define, configure, and maintain these pipelines. We developed the AutoPlan-PipelineCreator, a dosimetrist-facing Windows application that streamlines cohort identification, pipeline template selection, validation, and handoff to downstream automation utilities for structure generation, plan optimization, and DVH-based evaluation. The tool supports pipeline version control by persisting configurations and case selections as versioned artifacts, enabling systematic evaluation of configuration changes across large retrospective datasets.
Methods
AutoPlan-PipelineCreator was implemented as a Windows Forms application on .NET Framework 4.8. The workflow supports database-driven cohort queries, large-case triage in searchable/filterable grids, template and pipeline selection from configurable files, persistence of versioned pipeline JSON artifacts, ESAPI script configuration, and RapidCompare[1]-based evaluation of automatically generated plans. The tool was evaluated by generating and executing versioned pipeline configurations for two common VMAT use cases (prostate and head-and-neck) to enable large-cohort plan generation and comparative assessment of configuration changes using consistent case selection and scoring criteria.
Results
AutoPlan-PipelineCreator was used to define, iteratively refine, and validate 7 automated planning pipelines using 478 prostate and head-and-neck cases. Initial cohort selection and pipeline definition (including pre-processing) required ~2 days on average; subsequent refinement typically required several additional days, driven primarily by automated plan optimization and dose calculation updates. The tool enabled reproducible cohort creation via versioned pipeline and case-list artifacts, reduced manual transcription through automated identifier validation, and supported efficient large-cohort re-execution for comparative evaluation across pipeline versions.
Conclusion
AutoPlan-PipelineCreator reduces the time and specialized expertise required to operationalize automated planning pipelines by standardizing case identification, validation, and handoff to automated planning and RapidCompare-based evaluation in the Varian/ARIA environment, enabling repeatable and scalable assessment of pipeline changes. [1] https://doi.org/10.1002/acm2.14152