两种自动勾画系统勾画头部小体积危及器官的对比

目的:比较MANTEIA和RT-Mind两种软件自动勾画头部小体积危及器官（OAR）的准确性。方法:选取30例头部肿瘤患者的电子计算机断层扫描（CT）影像和核磁共振影像，将两套勾画系统软件自动勾画的实验组1和在勾画基础上手动修改的实验组2与医生手动勾画的对照组进行交叉指数系数（OI）、形状相似性系数（DSC）、杰卡德相似系数（J）和剂量偏差比较。结果:对于体积相对较大或CT值差异明显的头部器官（如脑干、晶体），OI、DSC、J值较高，剂量偏差较低；但对于小体积OAR（如视交叉、视神经、垂体）OI、DSC、J值较低，剂量偏差较高。P值分析发现:自动勾画A1组与手动勾画M组OAR的OI、DSC和J值比较差异均有统计学意义（P<0.05）。除晶体和内耳的OI值外，自动勾画B1组与手动勾画M组OAR的OI、DSC和J值比较差异均有统计学意义（P<0.05）。在自动勾画基础上，手动修改后，OI、DSC和J值都有提升，但与M组的OAR相比仍存在一定的差异。结论:通过软件自动勾画可以满足体积相对较大或CT值差异明显的头部器官放疗的临床需求，但对于头部小体积OAR，在临床上仍需要医生手动勾画。 【关键词】自动勾画；手动勾画；危及器官；头部肿瘤；剂量偏差

Comparison of two different systems for automatic segmentation of small-sized organs-at-risk in the head

Abstract: Objective To compare the accuracies of MANTEIA and RT-Mind in the automatic segmentation of small-sized organs-at-risk (OAR) in the head. Methods In the computed tomography (CT) images and magnetic resonance (MR) images of 30 patients with head tumors, the OAR were auto-segmented by MANTEIA and RT-Mind (experience group 1) and further manually modified (experience group 2), and also manually delineated by a doctor (control group). The overlap index (OI), Dice similarity coefficient (DSC), Jaccard similarity coefficient (J) and dose deviation were calculated and compared between groups. Results For head OAR with relatively greater volumes (such as brainstem) or obvious differences in CT values (such as lens), the OI, DSC and J value were higher, and the dose deviations were lower. However, for small-sized OAR (such as optic chiasm, optic nerve, pituitary), the OI, DSC and J value were lower, and the dose deviations were higher. P-value analysis showed the differences in OI, DSC and J value between experience group A1 and control group were statistically significant (P<0.05). Except the OI of the lens and inner ear, there were significant differences between experience group B1 and control group in OI, DSC and J value (P<0.05). After manual modification based on auto-segmentation, the OI, DSC and J value were further improved, but there still existed discrepancy as compared with segmentation results in control group. Conclusion Automatic segmentation with the above-mentioned systems can meet the clinical needs of radiotherapy for head organs with relatively greater volumes or obvious differences in CT values. However, for small-sized OAR, manual delineation is still needed in clinical practice.