Grading Evaluation Study Of Atlas-based Auto-segmentation Of Organs At Risk In Thorax

In radiotherapy,reducing radiation dose to organs at risk(OARs)as much as possible is one of the basic goals of radiation treatment planning.In clinical work,OAR delineation needs professional radiation oncologists to contour manually,which is timeconsuming,laborious,subjective,unrepeatable,and exists intra and inter-observer variability.Therefore,many auto-segmentation software has been developed and studied.At present,the accuracy of auto-segmentation software for different OARs has not been clear.Meanwhile,the evaluation methods have not been standardized.For this problem,this study graded comprehensively and accurately the feasibility on atlasbased auto-segmentation of thoracic OARs,and developed the grading standard.The main research contents of this paper are as below:(1)Grading evaluation study of atlas-based auto-segmentation of organs at risk in thorax.The Dice similarity coefficient(DSC),the difference of the Euclidean distance between centers of mass(ΔCMD),the difference of volume(ΔV),maximum Hausdorff distance(MHD)and average Hausdorff distance(AHD)between three autosegmented and manual contours were calculated.The atlas-based auto-segmentation strategies included single-atlas(Single),majority voting with 5 atlas matches(MV5)and simultaneous truth and performance level estimation(STAPLE)with 5 atlas matches(ST5).And the best auto-segmentation accuracy on thoracic OARs were graded into 3 levels according to each index.The results showed that grading evaluation of atlas-based auto-segmentation of thoracic OARs based on the DSC proved to be feasible and relatively more reliable.Forty patients with thoracic cancer were included in this study.The thoracic OARs auto-segmentation can be divided into three levels based on the DSC.This study recommended the clinical use of atlas-based auto-segmentation tools: the right lung,left lung,skin,heart and spinal cord can be automatically segmented,whereas the aorta,chest wall,trachea and pulmonary artery required to be manually modified after autosegmentation and the superior vena cava,esophagus,inferior vena cava,pulmonary vein and brachial plexus were not recommended for the automatic segmentation.(2)A novel specific grading standard study of auto-segmentation of organs at risk in thorax—subjective-objective-combined(SOC)grading standard.In our former proposed work,the grading standard was empirically developed by geometric indexes.Therefore,we proposed a novel SOC grading standard of auto-segmentation for each OAR in thorax.The development of the standard based on objective geometric evaluation and subjective evaluation.For the objective geometric evaluation,OAR auto-segmentation accuracy was graded by five geometric indexes of DSC,ΔCMD,ΔV,MHD and AHD.The grading results of them were compared with those of the corresponding geometric indexes in other two centers’ geometric objective methods.For the subjective evaluation,OAR auto-segmentation accuracy was also graded by the subjective evaluation standard proposed by us.Finally,based on the subjective evaluation standard and the five geometric indexes,the correspondence between the subjective evaluation level and the geometric index range was established for each OAR,which was the OAR-specific SOC grading standard in thorax.Combined with the easy-to-operate subjective evaluation standard and the DSC,the SOC grading standard evaluated the auto-segmentation accuracy of the thoracic OARs.Compared with the current geometric objective evaluation method and subjective evaluation method,the SOC grading standard represents a great improvement in accuracy comparison of different auto-segmentation software.