ABSTRACT
@#【Objective】To constructset-up errors distribution prediction model for pelvic tumors patients radiotherapy of Varian NovalisTX medical linear accelerator.【Methods】The IGRT set-up errors data of 30 pelvic tumor patients after radiotherapywere analyzed，and Gaussian Mixtures was used to construct the errors distribution prediction model and to solve the model parameters.【Results】The set-up errors is mainly concentrated in four central points of μ1~μ4. The coordinates of each central points shows that the offsets of vertical（-3.88~4.28）mm and Longitudinal（-2.41~1.54）mm were larger than that of lateral（-1.85~0.72）mm. The possibility of offsets clustering near μ2 and μ4（0.301，0.310）is greater than μ1 and μ3（0.190，0.196）. And the standard deviation of set- up errors can be up to 5.2 mm.【Conclusions】The Gaussian Mixtures Model made quantitative description and prediction analys is for the set- up errors distribution，which provide references for set-up errors control and planning target expansion of pelvic tumor radiotherapy in the absence of IGRT.