Room: AAPM ePoster Library
Purpose: predicting models of gamma passing rate (GPR) using a complexity index or artificial intelligence have been proposed for improving the efficiency of the patient-specific quality assurance (QA) processes. The current study aims to develop a convolution neural network (CNN) model for predicting GPR with 3D dose distribution and accumulated dose uncertainty potential (DUP) distribution.
Methods: hundred and thirty-five treatment plans of the prostate cancer were used in this study. The DUP distribution was generated by an in-house software that accumulated field edges weighted by a segmental monitor unit followed by Gaussian folding. The dose and DUP distributions on the detector element plane of these QA plans were used as input data for the predicting model. The GPR values were measured by ArcCHECK, and were used as output data for the predicting model. Our CNN model comprised nineteen layers and was trained for 200 epochs. The network was optimized using Adam. Five-fold cross-validation was applied to verify the predicting performance. The GPR values were predicted for twenty test cases, and the difference between the measured and predicted GPR values was evaluated.
Results: absolute error (MAE) between measured GPR and predicted GPR in the CNN model using the dose distribution was 2.12%, 2.53%, 2.15%, and 2.76% for 3%/3 mm, 3%/2 mm, 2%/3 mm, and 2%/2 mm tolerances, respectively. By adding the DUP distribution to the CNN model with the dose distribution, the MAE was improved to 1.84%, 2.39%, 1.86%, and 2.66% for 3%/3 mm, 3%/2 mm, 2%/3 mm, and 2%/2 mm tolerances, respectively.
Conclusion: CNN model with the dose and DUP distributions on the detector element plane enabled the prediction of GPR value with high accuracy. These findings could potentially help to omit the patient-specific QA measurements.
Quality Assurance, Intensity Modulation, 3D