【摘要】 目的建立基于增强CT的影像组学模型,探讨其鉴别卵巢浆液性囊腺瘤（SC）与交界性浆液性肿瘤（SBT）的诊断价值。方法回顾性分析经病理证实的49例卵巢SC患者及31例SBT患者的CT资料。由2名医师分别采用AK软件分析CT图像,勾画ROI,提取影像组学参数。对获得的影像组学特征参数进行多因素Logistic回归分析,建立预测模型;采用ROC曲线分析预测模型对卵巢SC与SBT的诊断效能。结果共提取396个影像组学参数,经降维处理后得到5个特征参数,分别为Percentile₁₀、Percentile₁₅、SA、LRHGLE_a90,o1及LRHGLE_a90,o7。2名医师提取参数的一致性良好（组内相关系数均>0.75）。以上述5个特征参数构建Radscore预测模型,在训练集中Radscore模型对鉴别诊断卵巢SC与SBT的AUC、敏感度、特异度分别为0.90、0.91、0.79,在测试集的AUC、敏感度、特异度分别为0.86、0.90、0.73。结论基于增强CT的影像组学模型能够有效鉴别卵巢SC与SBT。更多还原

【Abstract】 Objective To establish radiomic model based on enhanced CT, and to observe the value of the model for distinguishing benign and borderline serous tumors of ovary. Methods Data of CT imaging of 49 patients with ovary serous cystadenoma（SC） and 31 patients with serous borderline tumors（SBT） confirmed by pathology were retrospectively analyzed. AK software was used by 2 radiologists to delineate ROI of the tumors, and radiomic parameters were extracted. Then multiple Logistic regression was applied to identify optimal radiomic features and construct the prediction model. ROC curve was used to analyze the diagnostic efficacy of radiomic parameters and model on ovarian SC and SBT. Results A total of 396 image radiomics parameters were extracted, and 5 feature parameters were obtained after dimensionality reduction, namely Percentile₁₀, Percentile₁₅, SA, LRHGLEa_90,o1 and LRHGLE_a90,o7, respectively. The results of reproducibility analysis of 2 radiologists had good consistency（all intraclass correlation coefficient>0.75）. Radscore prediction model was constructed with the above 5 characteristic parameters, and the AUC, sensitivity and specificity of Radscore model for differentiating ovarian SC and SBT in the training set was 0.90, 0.91 and 0.79, while in the testing set was 0.86, 0.90 and 0.73, respectively. Conclusion Radiomic model based on enhanced CT can be used for identifying SC and SBT of ovary.更多还原