CT影像组学预测胃癌不同预后组织学分型的可行性

目的: 探讨CT影像组学在胃癌不同预后组织学分型预测中的应用价值。方法: 回顾性分析570例胃癌患者的临床资料，按组织学分型分为预后较差组和预后较好组，按照2 ∶1的比例随机分为训练组和验证组。采用软件ITK SNAP从CT图像中分割胃癌原发病灶。采用多因素分析筛选出与胃癌不同预后组织学分型最大相关性的临床特征，并以此构建临床特征模型。从所有分割图像中提取特征，再使用最大相关最小冗余(mRMR)、最小绝对收缩和选择算子(LASSO)回归和逐步回归筛选出有效特征，并以此建立影像组学标签。最后，拟合影像组学标签与临床特征，构建影像组学模型，并用受试者工作特征（ROC）曲线下面积（AUC）进行性能评估。结果： 提取了985个二维图像特征，包括一阶统计量、形状特征、纹理特征等。通过去冗余特征筛选，得到5个最有效特征构建影像组学标签。3个临床特征（年龄、性别、CT M分期）作为构建临床特征模型的参数。相较于临床特征模型和影像组学标签，影像组学模型的预测性能表现最佳，影像组学模型AUC在训练组和验证组分别为0.726 2(95%CI：0.676~0.776)，0.707(95%CI：0.632~0.781)。结论： CT影像组学在术前无创预测胃癌不同预后组织学分型方面具有一定的应用潜力。

The feasibility of CT-based radiomic nomogram for predicting histological classification of gastric cancer with different prognosis

Objective: To explore the application value of CT based radiomic nomogram in predicting histological classification of gastric cancer with different prognosis. Methods: A total of 570 patients with gastric cancer were analyzed retrospectively，and divided into poor prognosis group and good prognosis group. All the samples were randomly divided into training cohort and validation cohort in the proportion of 2 ∶1. Standardized CT images were segmented by radiologists with ITK SNAP software manually.The clinical features with the greatest correlation with histological classification of gastric cancer were screened by univariate analysis, and used as parameters to construct the clinical feature model. Features are extracted from all segmented images, and minimum Redundancy Maximum Relevance (mRMR), Least Absolute Shrinkage and Selection Operator (LASSO) regression and stepwise regression were used to screen the redundant radiomic features. The effective characteristics were used to establish radiomic signature.Finally, the clinical features closely related to histological classification were fitted, and radiomic nomogram was constructed.The performance of each model was evaluated by the area under ROC curve (AUC). Results: A total of 985 features were extracted from all the segmentation region, which included first order statistics , shape features, texture features, wavelet features and other filtering feature. After the screening of redundant features, five key features were obtained, which were used as the parameters of radiomic signature. Three clinical characteristics (age, sex, CT M) were selected as the parameters of the clinical feature model. Compared with other two models (clinical feature model and radiomic signature), the classification performance of radiomic nomogram is the best.The AUC of radiomic nomogram in training cohort and validation cohort was 0.726 2 (95%CI:0.676-0.776) and 0.707 0(95%CI:0.632-0.781). Conclusion: CT-based radiomic nomogram has certain potentiality in non invasive predicting different prognostis related to histological classification of gastric cancer before operation.