基于超声影像组学特征列线图模型术前鉴别早期与中晚期宫颈鳞癌

Nomogram model based on ultrasonic radiomics features for preoperative differentiation of early and mid-late stage cervical squamous cell carcinoma

目的 观察基于经阴道超声影像组学特征建立的列线图模型术前鉴别早期与中晚期宫颈鳞癌的价值。方法 回顾性收集经术后病理证实的227例宫颈鳞癌患者,利用3D-Slicer软件于术前经阴道声像图中勾画ROI,提取并经冗余性分析、最小绝对收缩和选择算子(LASSO)和10折交叉验证筛选影像组学特征,构建影像组学模型并得到Radscore评分;利用多因素logistic回归纳入Radscore及临床资料构建列线图模型。比较2个模型术前鉴别早期与中晚期宫颈鳞癌的受试者工作特征曲线下面积(AUC);评估列线图模型的校准度及临床收益。结果 最终纳入18个超声影像组学特征;以之构建术前鉴别早期与中晚期宫颈鳞癌的影像组学模型在训练集和验证集的AUC分别为0.839和0.744;联合年龄、流产次数及Radscore评分构建的列线图模型在训练集和验证集的AUC分别为0.882和0.773。DeLong检验结果显示,上述2模型在训练集的AUC差异有统计学意义(P<0.05)。Hosmer-Lemeshow检验显示,列线图模型在训练集和验证集的校准度均佳(χ^(2)=5.053、7.063,P均>0.05);决策曲线分析(DCA)显示其在0.01~1.00阈值区间净收益相对较大。结论 基于经阴道超声影像组学特征的列线图模型可于术前较好地鉴别早期与中晚期宫颈鳞癌。

Objective To explore the value of nomogram model based on transvaginal ultrasonic radiomics features for preoperative differentiation of early and mid-late stage cervical squamous cell carcinoma.Methods A total of 227 patients with cervical squamous cell carcinoma confirmed by postoperative pathology were retrospectively collected.3D-Slicer software was used to delineate lesion ROI on preoperative transvaginal ultrasound images.Imaging features within ROI were extracted,redundancy analysis,least absolute shrinkage and selection operator(LASSO)as well as ten-fold cross-validation were performed to screen important radiomics features.Then radiomics model was constructed,and Radscore was calculated.Finally a nomogram model was established combining with clinical data and Radscore using multivariate logistic regression.The area under the receiver operating characteristic curve(AUC)for preoperative differentiation of early and mid-late stage cervical squamous cell carcinoma of radiomics model and nomogram model were compared.Calibration and decision curve analysis(DCA)of nomogram model were assessed.Results Eighteen radiomics features were finally enrolled,and a radiomics model for preoperative differentiation of early and mid-late stage cervical squamous cell carcinoma was constructed.AUC of this model in training set and validation set was 0.839 and 0.744,respectively.A nomogram model was constructed combining with age,number of abortions and Radscore,with AUC of 0.882 and 0.773 in raining set and validation set,respectively.DeLong test showed that AUC of these 2 models were significantly different in training set(P<0.05).Hosmer-Lemeshow test indicated that the calibration degree of the nomogram model was good in both training set and validation set(χ^(2)=5.053,7.063,both P>0.05).Decision curve analysis(DCA)showed that the net benefit of this nomogram model was relatively high in threshold of 0.01—1.00.Conclusion Nomogram model established based on transvaginal ultrasonic radiomics features could effectively distinguish early and mid-late stage cervical squamous cell carcinoma preoperation.