Application Of Altas-Based Autosegmention Software In Esophageal Cancer

Objective: To evaluate the accuracy of atlas-based autosegmentation(ABAS)software when contouring critical organ-at-risk(OAR)in esophageal cancer,and assess the clinical feasibility of this technology.Methods: The CT images which included contoured OARs of 10 esophageal cancer patients treated with intensity-modulated radiotherapy(IMRT)were selected to establish ABAS reference atlas.From December 2015 to February 2016,10 new cases with stage III-IV esophageal cancer were randomized in this trial.The manual,single-template and multi-templates methods were adopted to contour their related OARs,respectively.Their morphological and dosimetric results were then compared.Wilcoxon signed-rank test was used for the comparison of their morphological indexes including the volume difference(△V%),the morphological consistency(Dice similarity coefficient and Overlap ratio)and the coordinates difference(△x,△y,△z).Correlation between the morphological consistency and OAR volumes was analyzed by Spearman method.And dosimetric endpoints of different coutour groups were compared using Friedmantest.Results: The average DSCvalues of single-template auto-contours and multitemplates auto-contours were0.83± 0.17 and 0.92 ±0.05,and the average ORvalues were 0.81± 0.17 and 0.92± 0.07,respectively.Significant differences were presented between them(P = 0.000,0.000).The coordinates displacements of multi-templates auto-contours in three directions were within 0.5cm,which is less than single-template ones.Spearman analysis found a positive correlation between DSC and OAR volumes.Thus the accuracy of single-template auto-contours is insufficient for spinal cord because of its small volume.The main dosimetric endpoints for the two auto-coutours were extremely similar to manual coutours.V20(%)for lung-all were 23.2 ± 9.0,22.4 ± 10.3 and 22.1 ± 9.9;Dmean(Gy)were 11.3 ± 4.0,11.1 ± 4.5 and 11.0 ± 4.3,respectively.For spinal cord,Dmax(Gy)showed 40.3 ± 4.8,38.2 ± 6.7,and 39.4 ± 5.3,respectively.And V30(%)for heart were 16.0 ± 14.7,15.8 ± 15.2 and 15.5 ± 14.4,respectively.The dose distribution showed no statistically significant difference between them(P > 0.05),and met the dose limit requirements of OARs for esophageal radiotherapy.Conclusion: The ABAS tool exhibits satisfactory accuracy for the OAR investigated in esophageal cancer with IMRT plans,which may potentially improve contour efficiency and reduce interoberserver variability.In addition,multi-templates method is better for clinical application.