论文部分内容阅读
Purpose:The accuracy of deformable image registration(DIR)algorithms could have a significant impact on treatment planning,especially for adaptive radiotherapy.Various DIR algorithms have been developed and integrated in commercial TPS.However,it's very difficult for a specific DIR algorithm to get universal clinical acceptable results from different patterns of organ deformation.In clinical routine,validation of the DIR algorithm is necessary.To evaluate the on-line adaptive radiotherapy potential for H&N cancer of RayStation,we gave a test on its integrated DIR algorithms using ten H&N cancer cases from the First Affiliated Hospital of Anhui Medical University.Methods:For each case,we took a CT(CT1)scan for treatment planning and a subsequent CT(CT2)for replanning after 3-4 weeks.OAR and PTV contours,including Brainstem,Spinal cord,Parotid Right,Parotid Left,Lens Right,Lens Left,were manually delineated by a radiation oncologist on both CT1 and CT2.ANACONDA mixed algorithm based on grey level and anatomic structure and MORFEUS algorithm based on biomechanical models,which are both integrated in RayStation,are used to deform CT1 and relevant contours onto CT2.The manually delineated contours and propagated corresponding contours deformed from CT1 on CT2 were evaluated by Dice coefficients for volume overlap,center of mass(CoM)displacement and Hausdorff distances(HDs)for max contour deformation displacement.We validated the performance of the two integrated algorithm using another multiparameter-optmized DIR algorithm called ELASTIX.Results:All the algorithms get a Dice coefficient of >0.80 for large organs with volumes >15.0 cm3.But contours for small organs with volume <1.0cm3,the Dice coefficient is between 0.5~0.8 and clinically need further validation.For small organ contours deformation,the Morfeus biomechanical algorithm shows better result relative to Hybrid mixed algorithm with Dice coefficient 0.14 higher by mean for both Lens Left and Lens Right.The brainstem shows the most significant deformation with max CoM of 6.3 mm and HDs of 11.6 mm.Compared to multipaprameter-optimized ELASTIX DIR algorithm,the Dice coefficients for ANACONDA and MORFEUS are lower by 0.42 and 0.01 separately than ELASTIX.Conclusions:The DIR algorithms in RayStation give an acceptable result for image registration in H&N adaptive radiotherapy.With the GPU-accelerated technique,it will be time-saving to realize on-line H&N radiotherapy.But for organs with volume <1.0 cm3,the results of deformable image registration need further validation.