Purpose: The objective of this study was to predict the failure points (gamma value > 1.0) and gamma passing rate (GPR) using a synthesized gamma maps of volumetric modulation arc therapy (VMAT) quality assurance (QA) with a deep convolutional generative adversarial network (GAN).
Methods: The fluence maps of two hundred and seventy VMAT beams for prostate cancer were measured by electronic portal imaging device, and were analyzed using gamma evaluation with 3%/2-mm, 2%/1-mm, 1%/1-mm, and 1%/0.5-mm tolerances. Created 270 gamma maps were divided into two datasets: training data for a model including 240 images and test data for evaluation including 30 images. Image prediction network of the fluence maps calculated by treatment planning system (TPS) to the gamma maps was created using GAN. The sensitivity, specificity, and accuracy for detecting failure points were evaluated from the measured and synthesized gamma maps. Additionally, the difference between the measured GPR (mGPR) and predicted GPR (pGPR) values calculated from the synthesized gamma maps was evaluated.
Results: Mean absolute error (MAE) between mGPR and pGPR was 0.5% for 3%/2-mm, 1.3% for 2%/1-mm, 2.6% for 1%/1-mm, and 2.7% for 1%/0.5-mm tolerances. The accuracy was 98.9% for 3%/2-mm, 96.9% for 2%/1-mm, 94.7% for 1%/0.5-mm, and 93.7% for 1%/0.5-mm tolerances. The sensitivity and specificity were the highest for 1%/0.5-mm and 3%/2-mm tolerances, which were 82.7% and 99.6%, respectively.
Conclusion: We developed the synthesized gamma map based patient-specific VMAT QA using GAN. The system is promising in quality assurance for radiotherapy because it showed a high performance and could identify the failure points.