Automation of radiotherapy treatment planning improves efficiency and consistency, while reducing planning time and errors. When adjustments are required to the automated plan there a numerous paths available to the dosimetrist that can lead to different outcomes impacting the resulting plan quality.

The objective of this study was to validate an Iterative Optimisation Engine (IOE) within an existing automated IMRT/VMAT planning framework. The IOE was designed to reduce remaining manual intervention within the automation framework through measuring and codifying best practice user intervention within a commercial planning system.

The IOE was developed for external beam IMRT/VMAT treatment planning on the Monaco V6.0.1 (Elekta AB, Stockholm, Sweden) Treatment Planning System. The IOE was built on an existing automation framework, utilizing the Application Programming Interface (API) to create completely automated treatment plans. For head and neck disease sites in Australia, a network of 40 centres evaluated automated treatment plans where users determined when manual intervention was required after automated planning in order for the clinician's approval. The modifications to automated plans were recorded, analysed, and codified into the API to remove the requirement for manual intervention. A subset of the automated plans were then retrospectively processed by the IOE with resulting plans being scored in three categories of 1) superior, 2) equivalent and 3) inferior based on DVH assessment with the original clinician approved plan as the baseline.

The automation framework generated 764 head and neck plans from January 1 2022 to August 1, 2023, of which 45% required manual intervention to achieve Dosimetric criteria. The IOE was inserted into the production system for head and neck optimisation, executed following the automated planning process. After being processed by the IOE, 86% of plans showed equivalent or superior coverage and maximum dose, and 95% of plans demonstrated equivalent homogeneity or improved homogeneity. Multi-target plans showed equivalent or improved target dose for 67% of intermediate dose targets and 39% of low dose targets when multiple targets were treated simultaneously. Analysis of organs at risk showed 38% of plans with reduced Parotid mean dose, 92% improved Larynx mean dose, 43% reduced Spinal Cord maximum dose, 57% decreased Brainstem maximum dose, 85% reduced Oral Cavity mean dose and 56% reduced Pharynx mean dose. Cases where improvement was negligible, it was observed that manual intervention after automation may have introduced some imbalances in the cost function weightings and prioritizations for organs at risk. Due to this potential imbalance, the IOE may have been limited in its ability to optimize and adjust certain plans effectively, thereby limiting its overall performance in this cohort of patients. In these cases, understanding a specific clinician's intent would help in the re-optimization process if required.

The addition of an IOE achieved a clinical improvement to target and OAR metrics in the assessed clinical plans. The automation framework will incorporate this work into clinical production to improve the overall effectiveness of the automated planning framework. Where iterative optimization is utilized, a reduction in manual intervention is anticipated which will aim to reduce instances where intervention unintentionally skews cost functions, limiting the flexibility of the IOC in its optimization process.