头颈部肿瘤放疗单模板自动轮廓勾画软件的应用  

Application of single-template atlas-based auto-segmentation in head and neck carcinoma

作　　者：胡江[1] 王明理[1] 徐林[1] 祁振宇[1] 

机构地区：[1]华南肿瘤学国家重点实验室中山大学肿瘤防治中心放疗科肿瘤医学协同创新中心,广东广州510060

出　　处：《中华肿瘤防治杂志》2015年第8期618-622,共5页Chinese Journal of Cancer Prevention and Treatment

摘　　要：目的探讨单模板自动轮廓勾画软件（atlas-based auto segmentation,ABAS）在头颈部肿瘤调强放疗（intensity modulated radiation therapy,IMRT）计划中器官轮廓勾画的准确性及其相关因素。方法选取中山大学肿瘤防治中心2013-06-01-2013-12-31收治的50例头颈部肿瘤患者资料进行回顾性分析,将1例体征参数接近中位平均值的患者CT平扫图像〔已完成14种危及器官（organs at risk,OARs）轮廓手动勾画并经临床确认〕导入ABAS软件作为通用模板,对软件自动生成与临床确认的器官轮廓计算形状相似性指数（dice similarity coefficient,DSC）,并统计相应勾画时间。依照患者体征（身高、体质量、头径、颈径和体表面积）对各器官DSC值进行等间距分组,采用SPSS 17.0进行单因素方差分析。对体征差异过大的患者分别基于通用模板和个体化模板进行自动勾画并通过配对样本t检验比较两者差异性。结果采用通用模板的大体积组和小体积组DSC均值分别为0.84±0.08和0.03±0.03,差异有统计学意义,t=27.806,P〈0.001;中体积组DSC值为0.80±0.12,与小体积组相比差异有统计学意义,t=22.848,P〈0.001。单因素方差分析结果证实,颞叶DSC值与患者头径体征存在相关性（颞叶-L：F=4.245,P=0.006,颞叶-R：F=4.264,P=0.006）,差异有统计学意义;腮腺DSC值与患者头径体征亦存在相关性（腮腺-L：F=2.934,P=0.043,腮腺-R：F=3.163,P=0.034）,差异有统计学意义。体征参数差异大患者分别采用通用模板和个体化模板进行自动勾画的脑干DSC值分别为0.75±0.08和0.82±0.06,差异有统计学意义,P=0.045。双侧颞叶DSC值分别为0.75±0.09、0.74±0.08和0.83±0.06、0.84±0.05,差异有统计学意义（颞叶-L：P=0.018,颞叶-R：P=0.041）。双侧腮腺DSC值分别为0.74±0.04、0.78±0.04和0.84±0.05、0.86±0.06,差异有统计学意义（腮腺-L：P=0.001,腮腺-R：P=0.001）。双侧内耳DSC值分别为0.67±0.11、0.66±0.12�OBJECTIVE To evaluate the accuracy of auto-contouring of normal tissues using a single-atlas-based au- to-segmentation （ABAS） software for head and neck cancer patients receiving intensity modulated radiation therapy （IM- RT） and to explore the correlative factors. METHODS The CT images of patient with normal figure （including 14 organs at risk （OARs） which had been clinically approved） were selected as the general atlases input for ABAS. A total of 50 head and neck cancer patients were retrospectively studied. Auto-contouring structures using ABAS were compared with those clinically approved by doctors and dice similarity coefficients （DSC） were calculated. The time spent in struc- ture delineations was also recorded. The possible factors that may affect the accuracy of auto-contouring including height, weight,head diameter,neck diameter and body surface area were analyzed by one-way ANOVA test using SPSS 17.0.In addition, individualized template was designed for special patients and was compared with the general template by Paired-sample T test. RESULTS The average DSC values of large-volume and small-volume group were 0.84±0.08 and 0.03±0.03, respectively. Independent sample T test showed that the difference was statistically significant （t= 27. 806, P〈0. 001）. The average DSC value of medium-volume group was 0.80 ± 0.12, the difference was statistically significant （t= 22. 848, P〈0. 001） compared to small-volume group. One-way ANOVA test showed the mean DSCs of temporal lobes and parotids had statistically significant relevance with the head diameter （temporal lobe-L： F= 4. 245, P= 0. 006, temporal lohe-R F= 4.264, P = 0. 006, parotid-L ： F = 2. 934, P = 0. 043, parotid-R.. F = 3. 163, P = 0. 034）. Compared to the general template, auto-segmentation based on the individualized template could significantly improve the mean DSCs of brain stem（0.75±0.08,0.82±0.06） ,temporal lobes（L：0.75±0.09,0.74±0.08 R：0.83±b0.06,0.84±0.05）, parotids （L：0.74±0.

关 键 词：自动轮廓勾画 单模板 形状相似性指数 调强放射治疗 

分 类 号：R739.91[医药卫生—肿瘤]