直肠癌术前放疗危及器官与临床靶区自动勾画的可行性研究

目的:探讨AccuContour软件及定制化自动勾画模型在直肠癌术前容积旋转调强放疗中临床靶区（CTV）和危及器官（OAR）自动勾画几何轮廓及剂量学各项参数精度，为临床应用提供依据。方法:回顾性选取133例已接受直肠癌术前容积旋转调强放疗的患者，随机分组，65例作为训练集，16例作为验证集，52例作为测试集，构建并训练自动勾画模型，将其导入AccuContour软件并自动勾画CTV和4个OAR，对比自动勾画与手动勾画在CTV和OAR几何轮廓的体积差异（ΔV]）、Dice相似性系数（DSC）、Jaccard系数（JAC）、敏感性指数（SI）、包容性系数（lncl）、质心偏差（DC）、Hausdorff距离（HD）等，以及自动勾画与手动勾画CTV和OAR在同一容积旋转调强计划中所受照射剂量学参数差异，从而评估自动勾画效果。结果:CTV的DSC值、JAC值、SI值、lncl值为:0.84±0.06、0.72±0.08、0.81±0.07、0.87±0.08，ΔV]值、DC值、HD值为:10.93%（4.56%, 15.37%）、5.03（3.27, 8.77） mm、15.03（15.00, 24.70） mm；OAR的DSC值、SI值、lncl值、JAC值、ΔV]值、DC值、HD值比较优劣顺序依次为:右股骨头、左股骨头、膀胱、小肠；自动勾画与手工勾画剂量学参数对比中，除膀胱V30、小肠Dmean、CTV D95的差异有统计学意义外（P<0.05），其余均无统计学意义（P>0.05）。结论:在直肠癌术前容积旋转调强放疗中，本研究所采用的自动勾画系统，对于CTV和OAR的自动勾画有一定准确性，为临床医生节省大量时间，提高工作效率。

Feasibility study of automatic segmentation of organs-at-risk and clinical target area for preoperative radiotherapy for rectal cancer

Abstract: Objective To explore the accuracy of AccuContour software and customized automatic segmentation model in automatically outlining the geometric contours of clinical target volume (CTV) and organs-at-risk (OAR) for the preoperative volumetric modulated arc therapy (VMAT) for rectal cancer, thereby providing a basis for clinical applications. Methods A total of 133 patients who have received preoperative VMAT for rectal cancer were enrolled retrospectively. Sixty-five cases of them were randomly selected for training set, 16 cases for validation set, and 52 cases for test set. The automatic segmentation model was established and trained, and then imported into AccuContour software for the automatic segmentations of CTV and 4 OAR. The automatic and manual segmentation results of CTV and OAR were compared in term of volume difference (ΔV]), Dice similarity coefficient (DSC), Jaccard coefficient (JAC), sensitivity index (SI), inclusion coefficient (Incl), centroid deviation (DC) and Hausdorff distance (HD). Moreover, the differences of dosimetric parameters between automatically and manually delineated CTV and OAR in the same VMAT plan were compared, so as to evaluate the performance of automatic segmentation. Results The DSC, JAC, SI, and Incl of CTV were 0.84±0.06, 0.72±0.08, 0.81±0.07, 0.87±0.08, respectively, and the ΔV], DC, and HD were 10.93% (4.56%, 15.37%), 5.03 (3.27, 8.77) mm, 15.03 (15.00, 24.70) mm, respectively. According to DSC, SI, Incl, JAC, ΔV], DC, and HD, the OAR segmentation performance from superior to inferior were as followed: right femoral head, left femoral head, bladder, small intestine. There were no statistical differences between automatic and manual segmentations in the dosimetric parameters (P>0.05) except bladder V30, small intestine Dmean, and CTV D95 (P<0.05). Conclusion In preoperative VMAT for rectal cancer, the automatic segmentation system used in the study has a high accuracy for the automatic segmentation of CTV and OAR, which saves a lot of time for clinicians and improves work efficiency.