Research On The Clinical Application Of Automatic Delineation Technology In Pelvic Radiotherapy

Objective:To evaluate and analyze the advantages and limitations of atlas-based autosegmentation(ABAS)tunderdosages in CTVs.The underdose volumes cannot be predicted by using hat outlined organs at risk(OAR)in radiotherapy of prostatic cancerMethod:Thirty patients with prostatic cancer who were treated with volumetric modulated arc therapy(VMAT)were retrospectively selected.The planning CT(pCT)images of the first and second course of same patient were inputted to ABAS software,and Prostate and General algorithms of the automatic delineating mode with single atlas were selected respectively.The planning CT(pCT)images of the first course were used as the reference images to realize automatic delineation for organs at risk of planning images of the second course.Dice similarity coefficient(DSC),overlap index(01),volume difference(Δ V)and the cumulative time of automatic delineation and manual modification were used,and the manual delineation results were adopted as reference to compare the results of automatic delineation of two algorithms.Results:Both DSC and 01,that were delineated automatically by two algorithms,of left and right hip bones and left and right femoral heads of same patient were higher than 0.8.Both DSC and 01 of bladder were higher than 0.7.And both DSC and 01 of left and right hip bones and bladder of Prostate algorithm were higher than those of General algorithm.While those of left and right femoral heads were similar as different algorithms.On the other hand,the value of ΔV of bony tissue was less than 0,which indicated that automatic delineation was insufficient in delineating this kind of tissue.Conclusion:There are differences between the results that are obtained from the applications of two differently automatic delineation algorithms and that of manual delineation.Prostate algorithm is better than General algorithm in delineating bladder and rectum.Moreover,the automatic delineation need to be calibrated and modified before it is applied to radiotherapy planning.Only in this way,it can truly realize the efficiency of ABAS delineation.Objective:To study the geometric accuracy and dosimetric accuracy of target volumes and organs at risk in postoperative endometrial carcinoma patients using atlas-based autosegmentation(ABAS)Method:Thirty-eight patients with endometrial cancer undergone radical resection were selected.Volumetric Modulated Arc Therapy(VMAT)were used for radiation therapy for these patients.The atlases for the ABAS were generated by using planning CT(pCT)images from 20 patients among them.the software was evaluated in the rest.The single(one to one)and multi-(nineteen to one)atlas mode in the software are selected to automatically delineate the CTVs and OARs on planning CT(pCT)images.Based on the non-reviewed auto-contours VMAT plans were constructed(referred to as ABAS plans)with the same beam configurations and dose constraints.The geometric differences[Dice similarity coefficient(DSC),Jaccard index(JI)]and dosimetric parameters were compared between two approaches of automatic delineation for CTVs and OARs.Results:The bony structures outlined for two automatic contouring methods is similar to the result of the manual contour(DSC>0.8).However,the CTVs and other OARs generated by a multi-atlas automatic segmentation(MAS)are closer to manual delineation than single Atlas method.According to the areas under ROC curves,the results of CTV and small intestine with multi-atlas segmentation are larger than that single-atlas automatic segmentation(SAS).The areas under the ROC curve of CTV with MAS and SAS are 0.732 and 0.624 respectively and those of small intestine are 0.704 and 0.562。The statistical differences of dosimetric parameters between automatic and manual segmentation are mainly in V95(manual outlining(MO)versus MAS:X2(2)=1.031,P=0.011;MO versus SAS:X2(2)=1.219,P=0.002)and Dmin(MO versus MAS:X2(2)=1.118,P=0.002;MO versus SAS:X2(2)=1.153,P<0.001)of CTV.Those of are demonstrated to CI(MO versus MAS:X2(2)=1.200,P=0.003;MO versus SAS:X2(2)=1.800,P<0.001)and HI(MO versus MAS:X2(2)=-1.267,P=0.002;MO versus SAS:X2(2)=-1.335,P<0.001)of PCTV and to Dmax of small intestine(MO versus MAS:X2(2)=-1.000,P=0.014;MO versus SAS:X2(2)=-1.625,P<0.001)Conclusion:Comparing the results of two algorithms in ABAS contouring the CTVs and OARs after radical operation of endometrial cancer,it was found that the multi-atlas segmentation is superior to the single-atlas segmentation in geometric accuracy and dosimetric accuracy.While,Editing and modifying of auto-segmentation generated by ABAS is required to avoid underdosages in CTVs.The underdose volumes cannot be predicted by using the similarity measures alone to evaluate the auto-segmentation algorithms...