中国癌症杂志 ›› 2020, Vol. 30 ›› Issue (3): 224-230.doi: 10.19401/j.cnki.1007-3639.2020.03.010

• 论著 • 上一篇    下一篇

应用CT纹理分析评价胰腺癌透明质酸含量

解添淞 1 ,马晓茜 2 ,童 彤 1 ,周正荣 1,3   

  1. 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 ;
    2. 复旦大学附属肿瘤医院病理科,复旦大学上海医学院肿瘤学系,上海 200032 ;
    3. 上海市闵行区肿瘤医院放射诊断科,上海 200240
  • 出版日期:2020-03-30 发布日期:2020-04-03
  • 通信作者: 周正荣 E-mail: zhouzr-16@163.com
  • 基金资助:
    上海市自然科学基金(15ZR1408000,12140901302,18140901200)。

Evaluating hyaluronan content of pancreatic cancer based on CT texture analysis

XIE Tiansong 1 , MA Xiaoxi 2 , TONG Tong 1 , ZHOU Zhengrong 1, 3   

  1. 1. Department of Radiology, Fudan University Shanghai Cancer Center, Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, China; 2. Department of Pathology, Fudan University Shanghai Cancer Center, Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, China; 3. Department of Radiology, Cancer Hospital of Shanghai Minhang District, Shanghai 200240, China
  • Published:2020-03-30 Online:2020-04-03
  • Contact: ZHOU Zhengrong E-mail: zhouzr-16@163.com

摘要: 背景与目的:肿瘤间质中大量聚集的透明质酸与胰腺癌的侵袭性和化疗抵抗密切相关,对其含量的评价将为胰腺癌的个体化治疗提供指导。纹理分析是一种基于定量特征无创评价肿瘤表型的图像分析方法。探讨CT纹理分析在评价胰腺癌透明质酸含量上的价值。方法:回顾性分析2015年6月—2015年12月复旦大学附属肿瘤医院经手术切除并且术前在复旦大学附属肿瘤医院行腹部CT检查的胰腺导管腺癌患者。应用免疫组织化学染色法评价肿瘤透明质酸含量。进一步使用3D-slicer软件在门静脉期手动逐层对肿瘤勾画感兴趣区(region of interest,ROI),使用基于Python的pyradiomics包对ROI提取56个纹理特征。使用组间相关系数评价特征的稳定性,并使用Mann-Whitney U检验选择出在高、低透明质酸含量组间存在显著差异的纹理特征。针对上述差异特征集,应用logistic逐步回归法构建模型,并使用受试者工作特征(receiver operating characteristic,ROC)曲线分析评价模型在鉴别高、低透明质酸含量的胰腺癌上的诊断效能。结果:纳入低透明质酸含量组13例,高透明质酸含量组17例,两组间未发现临床特征上的显著差异。纹理特征Volume、LeastAxis、Skewness、Energy在不同透明质酸含量的胰腺癌中存在显著差异。其中Skewness诊断效能最佳并被纳入模型,ROC曲线的曲线下面积为0.738,特异度为69.2%,灵敏度为76.5%。结论:CT纹理分析在评价胰腺癌透明质酸含量上有一定的应用价值。

关键词: 胰腺肿瘤, 纹理分析, 透明质酸

Abstract: Background and purpose: The increased accumulation of hyaluronan correlates with cancer aggressiveness and chemoresistance. Evaluating hyaluronan content may provide information to individualized management in patients with pancreatic cancer. As an image analysis approach, texture analysis could predict tumor phenotype based on quantitative features in a non-invasive manner. This study aimed to investigate the value of CT texture analysis in evaluating hyaluronan content of pancreatic cancer. Methods: Patients with pancreatic ductal adenocarcinoma confirmed by pathology after surgery were retrospectively enrolled in Fudan University Shanghai Cancer Center from Jun. 2015 to Dec. 2015. All patients underwent CT examination before resection. According to hyaluronan staining in immunohistochemistry, patients were divided into high and low hyaluronan group. Clinical characteristics were compared between them. On portal venous CT images, region of interest (ROI) was manually segmented to encompass whole tumor by 3D-slicer. The pyradiomics package based on Python was used to extract 56 texture features. The inter-class correlation coefficient was used to assess the reproducibility of texture features. The Mann-Whitney U test was applied to select valuable features exhibiting significant difference between two groups. Stepwise logistic regression was applied to develop a model discriminating high/low hyaluronan tumor. The diagnostic power was assessed by receiver operating characteristic (ROC) curve analysis. Results: There were 13 patients in low hyaluronan group and 17 patients in high hyaluronan group. There was no significant difference in clinical characteristics between two groups. Four texture features (Volume, LeastAxis, Skewness, Energy) exhibited significant difference between two groups. Eventually, Skewness was selected into logistic regression by two-sided stepwise method. The area under curve of model was 0.738, with a specificity and sensitivity of 69.2% and 76.5% respectively. Conclusion: CT-based texture analysis is valuable in evaluating hyaluronan content of pancreatic cancer.

Key words: Pancreatic neoplasm, Texture analysis, Hyaluronan