基于增强CT影像组学在宫颈癌根治性放疗疗效的研究

Study on the Curative Effect of Radical Radiotherapy for Cervical Cancer Based on Contrast-Enhanced CT Radiomics

目的 探讨基于增强CT的影像组学对宫颈癌患者根治性放疗疗效的二分类预测,进一步对其进行四分类预测,并建立四分类预测模型,为临床宫颈癌精准治疗提供基础支撑。方法 回顾性分析2016年7月至2021年7月川北医学院附属医院肿瘤放疗科收治的宫颈癌患者316例,行根治性放疗后评效结果为:完全缓解(CR)77例、部分缓解(PR)161例、疾病稳定(SD)71例和疾病进展(PD)7例。将纳入患者治疗前增强CT图像进行感兴趣区(ROI)勾画,并提取相关影像组学特征,然后采用自适应合成抽样法将数据平衡至644例。根据CR、PR、SD、PD四分类标签经过互信息、基于随机森林递归特征消除法得到最相关的影像组学特征,分别带入逻辑回归(LR)与支持向量机(SVM)进行模型训练、验证和测试,根据模型评价指标选出最优模型。绘制各模型的受试者工作特征曲线(ROC),并计算曲线下面积(AUC)。结果 经过特征筛选后,选出5个最相关的影像组学特征,对于LR四分类模型:测试集准确率为0.531,micor AUC值与macor AUC值分别为0.755、0.737;SVM四分类模型:测试集准确率为0.636,micor AUC值与macor AUC值分别为0.827、0.803。两种模型经过Delong检验P=0.046<0.05,AUC差值具有统计学意义。结论 经过严谨的处理过程后,基于增强CT的影像组学,不仅能够对宫颈癌根治性放疗疗效进行二分类预测,也能够对四分类进行一定的预测。SVM模型能够较好地对宫颈癌根治性放疗疗效的四分类预测(micor AUC=0.827)。

Objective To explore the dichotomic prediction of curative effect of radical radiotherapy for patients with cervical cancer based on enhanced CT imaging,and to further make a four-classification prediction,and establish a four-classification prediction model,to provide basic support for clinical precision treatment of cervical cancer.Methods A retrospective analysis was performed on 316 patients with cervical cancer admitted to the Department of Oncology and Ra-diotherapy,Affiliated Hospital of North Sichuan Medical College from July 2016 to July 2021,and the evaluation results after radical radiotherapy were as follows:There were 77 cases of complete responds(CR),161 cases of partial response(PR),71 cases of stable disease(SD)and progressive disease(PD)in 7 cases.Region of interest(ROI)was delineated on the enhanced CT images of the included patients before treatment,and the related image radiomics features were extracted.Then the data were balanced to 644 patients by adaptive composite sampling.According to CR,PR,SD and PD classification tags,the most relevant image radiomics features were obtained through mutual information and based on random forest recursive feature elimination method,which were respectively brought into logistic regression and support vector machine for model training,verification and testing,and the optimal model was selected according to model evaluation indicators.Results After feature screening,the 5 most relevant image radiomics features were selected.For the Logistic regression four classifi-cation model,the accuracy of test set was 0.531,micor AUC value and Macor AUC value were 0.755 and 0.737,respec-tively.Support vector machine four classification model:test set accuracy is O.636,micor AUC value and Macor AUC value are 0.827,0.803 respectively.The two models were tested by Delong test,P=0.046<0.05,and the Difference of AUC was statistically significant.Conclusion After rigorous treatment,imaging radiomics based on enhanced CT can not only predict the curative effect of radical radiotherapy for cervical cancer by dichotomies,but also predict the curative effect by dichotomies.SVM model can better predict the efficacy of radical radiotherapy for cervical cancer by four categories(Micor AUC=0.827).