A Quantitative Evaluation Method For Radiotherapy Plans Quality And Its Application In Clinical Rectal Cancer

The quality of radiotherapy plans may vary owing to differences in experience,habits,and other factors among physicists.Therefore,a quantitative plan evaluation method is very important in comparing different plans quality and guiding physicists in the design of a plan.This project aims to establish a quantitative evaluation method of radiotherapy plan based on comparison of two dose prediction models to provide guidance for clinical practice.The research content and results include two parts:(1)Comparison of dose prediction methods based on deep learning.With the continuous development of deep learning technology,great progress has been made in radiotherapy dose prediction.At present,there are many different dose prediction models with high accuracy for different tasks.In this project,two existing dose prediction models,C3D and HD UNet,were selected for comparison,and the model suitable for this project was selected.After preprocessing the rectal cancer case data,training and testing were performed,and random flip,rotation and translation operations were applied.The results showed that on the 15 test patients,both models achieved high dose prediction accuracy in the considered PTV and organs at risk.The MAE values of HD U-Net in PTV and left and right femoral heads were lower than those of C3D,and the mean absolute deviation values of HD U-Net in PTV dosimetry parameters D95%、Dmax、V107%and HI were lower than those of C3D,and only CI was slightly higher than that of C3D;Except for for Dmax of left femoral head,the mean absolute deviations of Dmax、V50、V45 and V35 of other organs at risk(small intestine,bladder and right femoral head)were not higher than those of C3D.Therefore,HD U-Net was selected as the dose prediction model in Chapter Ⅲ.(2)Quantitative evaluation methods of radiotherapy plans based on dose prediction.On the basis of the predicted dose based on HD U-Net,13 dose indicators of clinical concern for rectal cancer were selected,and weights and scoring functions were assigned according to their clinical importance.The scores of each indicator were summed as the final score of the plan.In addition,for organs with low scores,the radiotherapy treatment planning system DeepPlan was optimized,and the scores and dose distribution of the plans before and after optimization were compared.The test results of 15 rectal cancer cases showed that the average score of the optimized plan was significantly higher than that of the pre-optimized plan by 14.25%(t=-7.950,P<0.05);In addition,the dosimetric parameters(Dmax、V50)of the left femoral head and right femoral head in the optimized plan were significantly lower than those in the pre-optimized plan(t=2.450～12.238,P<0.05).In summary,this study successfully established a quantitative evaluation method for radiotherapy plan quality based on dose prediction.The effectiveness of the method was evidenced from the verification of 15 rectal cancer cases.