ESTRO 2022 - Abstract Book

S649

Abstract book

ESTRO 2022

Purpose or Objective Dose calculation during treatment planning in VMAT is a time-consuming process that can be reduced using the deep learning method. The purpose of this study is to evaluate the 3D dose distribution prediction accuracy from three different models. Materials and Methods Three models were designed with a different input data structure to train and test the model; 1) patient CT alone (PCT alone), 2) patient CT and generalized organ structure (PCTGOS), and 3) patient CT and specific organ structure (PCTSOS). The generative adversarial network model (GAN) was used as a core learning algorithm. The models were trained slice-by- slice using 46 VMAT prostate cancer plans, then predicted and evaluated the dose distribution from 8 independent plans. 3D gamma comparison with 3%,3mm, and percentage difference of DVH with criteria of Dmax, Dmean, D2%, D95%, and D98% were applied for model evaluation. Results The data training time with 500 Epoch took around 175 hours for 46 VMAT plans with all three models. Predicting 3D dose distribution for prostate VMAT takes approximately from 3.5 – 17.5 seconds per patient. PCTGOS model was the most reliable for predicting 3D dose distribution. The highest average 3D gamma passing rate (3%,3mm) was 80.51±5.94, and the average percentage difference of DVH was 6.01 ± 5.44% for PTV78, PTV60, PTV46, Bladder, Rectum, Left & Right femoral head.

Fig 1. The input CT image, ground truth dose distribution, and predicted dose distribution results of PCT alone model, PCTGOS model, and PCTSOS model.

Fig 2. Dose profile comparison between ground truth (blue line) and predicted (orange line) for 3 models: PCT alone model; PCTGOS model; and PCTSOS model. Conclusion This approach will accelerate the process by guiding and confirming the achievable dose distribution to reduce the replanning iterations while maintaining the plan quality.

PD-0737 Reducing plan complexity using the Aperture Shape Controller in VMAT head and neck treatments

A. Swinnen 1 , C. Wolfs 1 , M. Öllers 1 , A. Vaniqui 1 , F. Verhaegen 1

1 Maastro, Radiotherapy, Maastricht, The Netherlands

Purpose or Objective VMAT plans for head and neck tumors can be highly modulated. It has been demonstrated that more complex treatment plans lead to higher disagreement between the planned and delivered dose distribution [1]. The recently developed Aperture Shape Controller (ASC) by Varian Medical Systems is a leaf sequencer which controls the size and shape of the MLC in the Photon Optimizer algorithm in the Eclipse TPS. We aim to verify whether the ASC can reduce plan complexity and improve dose verification results (without comprising plan quality). Materials and Methods Six clinical head and neck treatment plans were reoptimized with five ASC settings ranging from “very low” to “very high” using the same optimization as the original patient plan. The different settings determine the strength of MLC modulation, for instance the ASC setting “very low” will result in highly modulated MLC movements with a lot of small-field segments.