四川大学学报（自然科学版）2025，Vol.62Issue(3)：661-668,8.DOI:10.19907/j.0490-6756.240234
基于人工神经网络重建6 MV医用直线加速器光子能谱
Reconstruction of 6 MV medical linear accelerator photon energy spectrum based on artificial neural network
摘要
Abstract
Artificial Neural Network(ANN)possess robust capabilities for solving linear equations and per-forming large-scale computations.In radiation therapy,the accuracy of incident photon beam energy spectrum directly affects the precision of dose calculations.This study employs an ANN to reconstruct the 6 MV medi-cal linear accelerator photon spectrum by inputting depth dose data.Initially,a training spectrum was calcu-lated using empirical formulas,and corresponding training depth doses were obtained by simulating monoen-ergetic photon depth doses with EGSnrc/DOSXYZnrc.The loss function was defined as the mean square er-ror between the reconstructed spectrum and the training spectrum,and the model was pre-trained accord-ingly.Subsequently,for iX,600C,and Primus linear accelerators under a 10 cm×10 cm reference field,as well as for an XHA600D linear accelerator under 4 cm×4 cm,10 cm×10 cm,and 20 cm×20 cm fields,the test spectra were analyzed based on phase-space files generated by EGSnrc/BEAMnrc,and test depth doses were simulated using EGSnrc/DOSXYZnrc.The loss function was modified to the mean square error be-tween the depth dose corresponding to the reconstructed spectrum and the test depth dose,followed by fine-tuning of the model.The results indicate that the reconstructed spectra using the ANN not only matched the peak position and full width at half maximum of the test spectra but also achieved a root mean square error of less than 0.51%.Under a 3 mm/3%tolerance,the γ passing rate for the depth dose corresponding to the re-constructed spectrum was over 98%.These findings validate the effectiveness of the ANN-based photon beam spectrum reconstruction method.关键词
直线加速器/人工神经网络/光子能谱/深度剂量Key words
Linear accelerator/Artificial neural network/Photon energy spectrum/Depth dose分类
核科学引用本文复制引用
李天航,刘伟,王诗露,吴章文,勾成俊..基于人工神经网络重建6 MV医用直线加速器光子能谱[J].四川大学学报（自然科学版）,2025,62(3):661-668,8.基金项目
国家重点研发计划(2016YFC0105103) （2016YFC0105103）
辐射物理及技术教育部重点实验室开放课题(2023SCURPT06) （2023SCURPT06）
