UNASSIGNED: In this retrospective study, 131 patients who underwent surgical resection and diagnosed HAE in pathology were included (bioactive, n=69; bioinactive, n=62). All patients were randomly assigned to the training cohort (n=78) and validation cohort (n=53) in a 6:4 ratio. The gross lesion volume (GLV), perilesional volume (PLV), and gross combined perilesional volume (GPLV) radiomics features were extracted on CT images of portal vein phase. Feature selection was performed by intra-class correlation coefficient (ICC), univariate analysis, and least absolute shrinkage and selection operator (LASSO). Radiomics models were established by support vector machine (SVM). The Radscore of the best radiomics model and clinical independent predictors were combined to establish a clinical radiomics nomogram. Receiver operating characteristic curve (ROC) and decision curves were used to evaluate the predictive performance of the nomogram model.
UNASSIGNED: In the training cohort, the area under the ROC curve (AUC) of the GLV, PLV, and GPLV radiomic models was 0.774, 0.729, and 0.868, respectively. GPLV radiomic models performed best among the three models in training and validation cohort. Calcification type and fibrinogen were clinical independent predictors (p<0.05). The AUC of the nomogram-model-based clinical and GPLV radiomic signatures was 0.914 in the training cohort and 0.833 in the validation cohort. The decision curve analysis showed that the nomogram had greater benefits compared with the single radiomics model or clinical model.
UNASSIGNED: The nomogram model based on clinical and GPLV radiomic signatures shows the best performance in prediction of the bioactivity of HAE. Radiomics including perilesional tissue can significantly improve the prediction efficacy of HAE bioactivity.
■在这项回顾性研究中,131例接受手术切除并在病理上诊断为HAE的患者(生物活性,n=69;无生物活性,n=62)。所有患者以6:4的比例随机分配到训练队列(n=78)和验证队列(n=53)。总病变体积(GLV),周围体积(PLV),并在门静脉期CT图像上提取大体合并病灶体积(GPLV)影像组学特征。通过类内相关系数(ICC)进行特征选择,单变量分析,和最小绝对收缩和选择运算符(LASSO)。通过支持向量机(SVM)建立影像组学模型。将最佳影像组学模型的Radscore与临床独立预测因子相结合,建立临床影像组学列线图。使用受试者工作特征曲线(ROC)和决策曲线来评估列线图模型的预测性能。
■在培训队列中,GLV的ROC曲线下面积(AUC),PLV,和GPLV影像组学模型分别为0.774、0.729和0.868。GPLV影像组学模型在训练和验证队列中的三个模型中表现最好。钙化类型和纤维蛋白原是临床独立预测因素(p<0.05)。基于列线图模型的临床和GPLV放射组学特征的AUC在训练队列中为0.914,在验证队列中为0.833。决策曲线分析表明,与单一影像组学模型或临床模型相比,列线图具有更大的优势。
■基于临床和GPLV放射组学特征的列线图模型在预测HAE的生物活性方面表现最佳。包括病灶周围组织的放射组学可以显着提高HAE生物活性的预测功效。