体部立体定向放射治疗摆位误差分布预测模型的构建OA

Objective To construct a prediction model for the setup error distribution of patients undergoing stereotactic body radiation therapy(SBRT).Methods The setup error data of 45 patients with SBRT who were treated with the Varian Vital Beam(SN3546)medical linear accelerator in the Department of Oncology Radiology of the Affiliated Hospital of Xuzhou Medical University from January 2019 to December 2022 were selected as the research objects.The error distribution prediction model was constructed by using the variational inference Bayesian Gaussian mixture model to solve the model parameters.Results The setup errors in the translation direction mainly concentrated in the three centers(μ1 to μ3)directions.The average offsets were relatively large in the front-back(-3.45 to 4.84 mm)and head-foot directions(0.22 to 5.67 mm),while the average offsets were relatively small in the left-right direction(0.14 to 0.79 mm).The possibility of offset of the setup error towards the center point of μ3 was the greatest,and the probability of the error center was 0.54.The possibility of offset towards μ1 was the smallest,and the probability of the error center was 0.20.The maximum variance in the translation direction was 1.72 mm.The setup errors in the rotation direction mainly concentrated in the directions of the four center points(μ1 to μ4).The average offsets in the tilt direction(-0.50° to 0.15°)and the rotation direction(-0.52° to-0.10°)were relatively larger than those in the rotation direction(0.03° to 0.25°).The possibility of the setup error offsetting towards the μ3 center point was the greatest,and the probability of the error center was 0.69.The offset possibilities of the remaining three center points were not much different.Compared with the translation direction,the variance values of each center in the rotation direction were larger except for the data of the μ2 center point(-0.52°),indicating that the data in the translation direction was more concentrated.Conclusion The variational inference Bayesian Gaussian mixture model can quantitatively describe and predict the setup error data,providing a reference for exploring the distribution regularity of SBRT setup errors.