Room: ePoster Forums
Purpose: The purpose of this study is to establish a prediction model for pretreatment dose verification of volumetric-modulated arc therapy (VMAT) using machine learning.
Methods: 248 cases of clinical VMAT were selected and pretreatment dose verification was carried out by a three-dimensional diode array. Thirty-four features were extracted from each patient's VMAT plan, which can reflect the beam parameters and the beam complexity of each plan. 198 cases were used to train the model and 50 cases were used for verification. Lasso regularized Poisson regression algorithm was used to train, and a prediction model was established to predict gamma passing rate (3%/3 mm criterion) through the characteristics of VMAT plan.
Results: The maximum deviation between the predicted and the measured gamma pass rate was within 5%. 92% of all prediction deviations was less than 3%, and 70% of all prediction deviations was less than 2%. The prediction model consists of eight features, and the most important features are MCSv, leaf gap 10 to 15 mm and mean AAV.
Conclusion: By extracting the features of the VMAT plan complexity and parameters, the gamma pass rate of pretreatment dose verification of VMAT can be effectively predicted. This method provides a new way to optimize the process of pretreatment dose verification for VMAT QA.
Funding Support, Disclosures, and Conflict of Interest: This research work was supported by National Natural Science Foundation of China (Grant No. 81472807).