A Method Of Evaluating Treatment Plans Based On Inter-Fractional CBCT Images And Its Application In Adaptive Radiotherapy Of Cervical Cancer

The radiation therapy process takes about 30 days of fractionated irradiation to deliver the total doses per treatment plan.If inter-fractional anatomical and physiological changes in the process were neglected,radiation dose errors can accumulate,leading to possible failure of the treatment in the end.To circumvent this problem,radiotherapy community has proposed the so-called "Adaptive Radiotherapy(ART)" strategy that makes timely adjustment to the treatment plans in response to such anatomical changes.However,the current implementation of Cone Beam CT(CBCT)-based inter-fractional ART process takes 20 to 30 minutes,thus squeezing resources out of already time-pressed Radiation Oncology departments which choose to deploy ART only when clinical benefits are apparent.Therefore,to date,ART urgently needs a fast yet accurate method that evaluates treatment plans based on inter-fractional CBCT images,keeping in mind of three challenges:(1)The quality of CBCT images is inferior to that of CT images for diagnostic purposes,and there is not much research on automatic segmentation based on CBCT images,resulting in poor speed and accuracy in delineation of target areas and organs at risk;(2)The CBCT images contains scattering artifacts and cavity artifacts,which affect the accuracy of dose calculation;(3)Accurate and fast adaptive radiotherapy treatment plan evaluation methods rely on new radiation dose calculation methods and software technologies.To address these issues,this study aimed to propose and validate an ART treatment plan evaluation method based on inter-fractional CBCT images.The main research and achievements of this study include:(1)Research on image registration methods based on the CBCT and CT images.This study adapted a number of methods including a rigid-registration process based on the ITK open-source library,a deformable registration algorithm based on B-spline algorithm,an AI-based cascade network deformable registration(MSnet)algorithm,and an ROI registration algorithm guided by structural information.The feasibility of the registration method was systematically evaluated from the perspective of image quality,contour propagation accuracy,and dose analysis.The MSnet was found to significantly improve registration efficiency(0.15s)while achieving the same accuracy as B-spline registration;B-spline registration could be used as a supplement to MSnet models with poor generalization,improving the generalization and robustness of deformable registration in this study;ROI registration could improve the contour propagation accuracy of CTV and PTV,with Dice coefficients of 0.867 and 0.917 for CTV and PTV,respectively;The gamma pass rate of 1mm/1%was more than 99%for the the dose distribution calculated using deformable CT images.(2)Research on generating sCT images based on CycleGAN.This study proposed an improved CycleGAN model and image preprocessing process to achieve sCT images from CBCT images,and verified the feasibility of the sCT generation method from the perspective of image quality and dosimetric parameters.Compared to CBCT images,the quality of sCT images was found to be significantly improved,the value of mean absolute error(MAE)was 43.98±10.74,the value of root mean square error(RMSE)was 117.58±28.22,the value of peak signal-to-noise ratio(PSNR)was 22.23 ± 2.61,and the value of structural similarity(SSIM)was 0.685 ±0.055.The contour accuracy of auto-segmentation based on sCT images was found to be superior to that of deformation registration propagated contours.As for the dose distribution calculated with sCT images,the gamma pass rate of 1mm/1%was found to be greater than 99%,and the errors of dosimetric parameters(D2 and Dmean)were less than 0.3%.(3)Treatment plan evaluation based on inter-fractional CBCT images.The deformable image registration and sCT generation methods have been integrated into a dose QA software system developed previously by our group,ArcherQA,to implement plan evaluation and dosimetric parameter analysis within 3-5 minutes.This study has embedded these algorithms into the ArcherQA Adapted Check module to allow for clinical tests involving 8 postoperative cervical cancer patients.Test results were found to reduce the error of the original plan from 4.3%to 2.6%.Results from this study suggest that,by overcoming a number of technological challenges,we have proposed and tested a treatment plan evaluation method based on inter-fractional CBCT images that meets restrict clinical requirements in terms of accuracy and speed,thus making available a quantitative analysis tool for dosimetric differences caused by anatomical changes in ART.