中国癌症杂志 ›› 2020, Vol. 30 ›› Issue (6): 468-474.doi: 10.19401/j.cnki.1007-3639.2020.06.010

• 论著 • 上一篇    下一篇

MSCT征象联合纹理分析在预测胸腺上皮性肿瘤WHO简化病理分型中的价值

任采月,王升平,张盛箭,彭卫军   

  1. 复旦大学附属肿瘤医院放射诊断科,复旦大学上海医学院肿瘤学系,上海 200032
  • 出版日期:2020-06-30 发布日期:2020-07-16
  • 通信作者: 张盛箭 E-mail: zhangshengjian@yeah.net

The value of MSCT signs combined with texture analysis in preoperatively predicting WHO simplified pathological classification of thymic epithelial tumors

REN Caiyue, WANG Shengping, ZHANG Shengjian, PENG Weijun   

  1. Department of Radiology, Fudan University Shanghai Cancer Center; Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, China
  • Published:2020-06-30 Online:2020-07-16
  • Contact: ZHANG Shengjian E-mail: zhangshengjian@yeah.net

摘要: 背景与目的:胸腺上皮性肿瘤(thymic epithelial tumor,TET)是前上纵隔最常见的原发肿瘤,其组织学分型是判断预后的独立危险因素。探讨和比较术前多层螺旋计算机断层扫描(multi-slice spiral computed tomography,MSCT)征象及基于CT图像的纹理分析在预测TET世界卫生组织(World Health Organization,WHO)简化病理分型中的价值。法:回顾并分析2011年1月—2018年6月复旦大学附属肿瘤医院手术后经病理学检查证实为TET、且经免疫组织化学确定组织分型的120例患者的增强CT图像及临床资料,并根据WHO简化病理分型分为低危组(A1、AB、B1型胸腺瘤)及高危组(B2、B3型胸腺瘤及胸腺癌)。评估及记录每例TET患者的CT征象。选用纵隔窗CT图像进行病灶分割及纹理特征的提取。采用R软件套索(least absolute shrinkage and selection operator,Lasso)模型进行特征筛选及模型建立。采用受试者工作特征 (receiver operating characteristic,ROC)曲线和曲线下面积(area under curve,AUC)评价模型的预测效能。不同预测模型之间诊断效能的比较采用DeLong检验。结果:120例TET患者中低危组61例,高危组59例,分别记录及提取了每例患者CT图像的11种CT征象和14种纹理特征。分别建立了以MSCT征象、纹理参数以及两者联合为基础的CT预测模型、纹理预测模型以及联合预测模型,3个模型在预测低危组及高危组TET的AUC分别为0.78、0.80、0.88,灵敏度分别为86.4%、88.1%和93.2%,特异度分别为60.7%、65.6%和67.2%,准确率分别为70.8%、74.2%和80.0%。DeLong检验结果显示,联合预测模型在预测TET恶性程度中的效能及准确率最高(P<0.05)。结论:MSCT检查是TET首选的影像学检查方法,纹理特征比肉眼评估得到的CT征象更能反映TET的微观异质性程度,CT征象与纹理特征相结合在预测TET恶性程度中更有优势,可为临床制订治疗方案及评估预后提供更全面及准确的依据。

关键词: 胸腺上皮肿瘤, WHO病理分型, 多层螺旋计算机断层扫描, 纹理分析

Abstract: Background and purpose: Thymic epithelial tumor (TET) is the most common primary tumor in the anterior superior mediastinum with considerable histologic heterogeneity. This study aimed to compare the value of preoperative multi-slice spiral computed tomography (MSCT) signs and texture analysis based on computed tomography (CT) images in predicting World Health Organization (WHO) simplified pathological classification of TET. Methods: A total of 120 patients with pathologically confirmed TET in Fudan University Shanghai Cancer Center from Jan. 2011 to Jun. 2018 were retrospectively analyzed. According to the WHO simplified pathological classification, patients were divided into low-risk group (A1, AB, B1 thymoma) and high-risk group (B2, B3 thymoma and thymic adenocarcinoma). CT signs of each TET patient were evaluated and recorded. Texture feature extraction was performed on CT images of mediastinal window. Predictive models based on CT signs, texture features and the combination of them were developed with the least absolute shrinkage and selection operator (Lasso) regression. The predictive effectiveness of the model was evaluated by receiver operating characteristic (ROC) curve and compared with the area under curve (AUC). DeLong test was used to compare the diagnostic effectiveness of different prediction models. Results: A total of 120 patients with TET were enrolled in this study. Totally 11 CT signs and 14 texture features of each patient were recorded and extracted. CT model, texture model and the combination model were established. The AUCs of the three models in predicting low-risk group and high-risk group TET were 0.78, 0.80 and 0.88, respectively. The sensitivity was 86.4%, 88.1% and 93.2%, respectively. The specificity was 60.7%, 65.6% and 67.2%, respectively. And the accuracy was 70.8%, 74.2% and 80.0%, respectively. DeLong test showed that the combination model held the highest predictive efficiency (P<0.05). Conclusion: MSCT examination is the preferred imaging examination method for TET. Texture features can better reflect the micro-heterogeneity of TET than those obtained by naked eye evaluation. The combination of CT signs and texture features has more advantages in predicting the malignant degree of TET, and provides more comprehensive and accurate basis for clinical treatment and prognosis evaluation.

Key words:  , Thymic epithelial tumor, WHO pathological classification, Multi-slice spiral computed tomography, Texture analysis