Clinical Evaluation of an Automated Treatment Planning Framework for Radiation Oncology

Abstract

<h3>Purpose/Objective(s)</h3> Automation of radiotherapy treatment planning increases consistency, provides a safety net around error prevention, and reduces planning time. The objective of this study was to validate a scalable framework to automate the creation of VMAT and IMRT treatment plans using commercial planning systems. <h3>Materials/Methods</h3> A total of 35 centers across Australia were used for the study and the disease sites consisted of Urinogenital, Lung, Breast, Head and Neck, Neurological, GI and Gynecological. A scalable framework for external beam treatment planning was implemented for two commercial treatment planning systems (TPS), Eclipse (Varian Medical Systems, Palo Alto, CA) and Monaco (Elekta AB, Stockholm, Sweden). The framework received CT images, target volumes, normal tissue structures, and prescription data from dosimetrists. Information was parsed into the system, validated against configuration settings, the desired treatment protocol template selected, optimization parameters adjusted, and dose calculation completed. The framework controlled the TPS through the Application Programming Interface (API). The dosimetry team evaluated each automated plan by comparison of the plan metrics to the prescription and assessment of the dose distribution. The plans were scored in three areas: 1) clinician approved with no intervention, 2) minimal intervention with minor adjustments, and 3) manual replanning. <h3>Results</h3> Since Aug. 2021, 4972 plans (36% of total plans) have been successfully processed across 35 treatment centers with 111 clinicians. 487 plans were rejected by the framework due to error. The system automated plans for GUI (1822), Lung (629), Breast (745), Head and Neck (65), Neurological (96), GI (568) and Gynecological (163). Out of these plans, 23% were clinician approved with no user intervention, 41% required minimal intervention to achieve clinician approval, and 36% required additional optimization structures and/or cycles of optimization to achieve clinician approval. The average time between clinician target volume delineation and plan approval reduced by 25% from 1.6 to 1.2 days. In Feb. 2022, the framework successfully processed 970 plans (41% of total plans), rejecting 126 (5%) due to error. Evaluation between systems showed 27% of Eclipse plans and 43% of Monaco plans were processed successfully. Of these plans, 19% (Eclipse) and 23% (Monaco) required no intervention while 25% (Eclipse) and 46% (Monaco) required minimal intervention. Analysis of dosimetrist productivity showed an increase of 16% since Aug. 2021 while the average time saved per plan was 85 minutes. Analysis of adverse and near- miss dosimetry related events showed a 6% reduction in reported incidents. <h3>Conclusion</h3> The framework for automated planning achieved a 60% clinician approval rate for plans with minimal or no intervention. Automation can result in more consistent planning and timesaving's for expert operators. A 6% reduction in recorded dosimetry related errors was achieved.