Research project

Knowledge-based planning for head and neck radiotherapy: an evaluation of the impact of RapidPlan on the quality and efficiency of treatment planning

Radiotherapy Physics
Susannah Leah
Training location
Norfolk and Norwich University Hospitals NHS Foundation Trust

Inverse-planning techniques such as intensity-modulated radiotherapy (IMRT) and volumetric-modulated radiotherapy (VMAT) enable highly conformal dose distributions; however, the trial-and-error approach required in the optimisation process can be time consuming and subjective. RapidPlan (Varian Medical Systems, Palo Alto, CA) is a knowledge-based planning solution able to predict dose-volume histograms for a given patient using a model built from previous patient plans.

The main objective of this work was to develop and validate a RapidPlan model for head and neck bilateral-nodes treatment planning which can produce plans that are comparable or superior to manually-produced, clinical plans. The secondary aim was to investigate the optimal use of RapidPlan ahead of implementation into local practice. A cohort of previous patient plans were used to train a RapidPlan model. In order to maximise model quality, 58% of the plans within the training set were refi ned. Data from clinical and model-generated plans were compared to test whether use of the model can improve plan quality with respect to local planning objectives. Model-generated plans obtained using the same optimisation and dose calculation algorithms were acceptable with respect to mandatory dose-volume objectives and had reduced parotid and submandibular gland mean doses.

However, the target coverage was slightly reduced. Model-generated plans obtained using updated algorithms met all optimal target coverage objectives, and were comparable or superior to the original clinical plans with respect to target and organ-at-risk planning objectives. It was not necessary to match submandibular glands fully overlapping with the target volume to the model structure trained from the combined volume of both glands.

Furthermore, different methods for bilateral parotid gland dose optimisation were built into the model. By revising the method selected it was possible to achieve greater dose reduction to the parotid and submandibular glands whilst still achieving optimal target coverage objectives. A RapidPlan model has been developed which can automatically produce VMAT plans for bilateral-nodes, head and neck cancer patients which are comparable or superior to historic manually-produced clinical plans. However, user involvement was found to be desirable when deciding which normal tissue volumes to match with the model.

Last updated on 4th October 2022