Improving the accuracy of pencil beam dose calculation of intensity-modulated proton therapy for lung cancer patients using deep learning / 中华放射肿瘤学杂志

ABSTRACT

Objective:Proton pencil beam (PB) dose calculation can achieve rapid dose calculation, whereas it is inaccurate due to the approximation in dealing with inhomogeneities. Monte Carlo (MC) dose calculation is recognized as the most accurate method, but it is extremely time consuming. The aim of this study was to apply deep-learning methods to improve the accuracy of PB dose calculation by learning the difference between the MC and PB dose distribution.Methods:A model which could convert the PB dose into the MC dose in lung cancer patients treated with intensity-modulated proton therapy (IMPT) was established based on the Hierarchically Densely Connected U-Net (HD U-Net) network. PB dose and CT images were used as model input to predict the MC dose for IMPT. The beam dose and CT images of 27 non-small cell lung cancer patients were preprocessed to the same angle and normalized, and then used as model input. The accuracy of the model was evaluated by comparing the mean square error and γ passing rate (1 mm/1%) results between the predicted dose and MC dose.Results:The predicted dose showed good agreement with MC dose. Using the 1 mm/1% criteria, the average γ passing rate (voxels receiving more than 10% of maximum MC dose) between the predicted and MC doses reached (92.8±3.4)% for the test patients. The average dose prediction time for test patients was (6.72±2.26) s.Conclusion:A deep-learning model that can accurately predict the MC dose based on the PB dose and CT images is successfully developed, which can be used as an efficient and practical tool to improve the accuracy of PB dose calculation for IMPT in lung cancer patients.