应用多参数MRI放射组学诺莫图预测宫颈癌淋巴结转移的研究  

Prediction of Cervical Cancer Lymph Node Metastasis Using Multiparametric MRI Radiomics Nomogram

作　　者：郑逸群 姜增誉[1,2] 张素鑫 李海燕[1] 李丽珍[1] 梁赟 董雨萌 何生[1,2] ZHENG Yi-qun;JIANG Zeng-yu;ZHANG Su-xin;LI Hai-yan;LI Li-zhen;LIANG Yun;DONG Yu-meng;HE Sheng(College of Medical Imaging,Shanxi Medical University,Taiyuan 030001,Shanxi Province,China;Department of Radiology,the First Hospital of Shanxi Medical University,Taiyuan 030001,Shanxi Province,China)

机构地区：[1]山西医科大学医学影像学院,山西太原030001 [2]山西医科大学第一医院放射科,山西太原030001

出　　处：《中国CT和MRI杂志》2025年第3期147-150,共4页Chinese Journal of CT and MRI

基　　金：山西省重点研发计划(201803D31004);山西省基础研究计划(202303021221214);山西省四个一批医学科研计划(2023XM031,2021074)。

摘　　要：目的旨在评估MRI影像组学参数在预测宫颈癌患者淋巴结转移中的应用价值,结合临床信息建立模型,以期实现术前对宫颈癌淋巴结转移状态的有效预测。方法回顾性分析2018年10月至2023年9月期间,在山西医科大学第一医院接受盆腔MRI检查并经手术病理确认的171例宫颈癌患者,这些患者中既有淋巴结转移也有未发生转移的情况。按照7:3的比例随机分配为训练集(n=120)和测试集(n=51)。通过多序列MRI图像手动划定病灶感兴趣区域(ROI),并运用Python程序从中提取特征值。采用Pearson相关系数与LASSO回归相结合的方法进行特征筛选及降维处理。利用SPSS 26.0软件完成连续型变量及分类数据的统计分析工作。借助MedCalc软件生成受试者操作特征(ROC)曲线来评价不同模型识别宫颈癌淋巴结转移的能力。结果研究发现,在训练集中,仅基于临床信息构建的模型、影像组学模型以及综合模型其曲线下面积(AUC)分别为0.759、0.828和0.848;而在独立验证队列中,这三个数值依次为0.809、0.806和0.848,显示出综合模型具备最优性能。此外,通过绘制列线图及校准曲线观察到预测概率与实际观察结果之间存在良好一致性。决策曲线分析进一步证实了该综合模型具有显著的临床实用性。结论将多参数MRI影像组学特征与传统临床指标相结合所开发出的新模型,在辅助医生于手术前准确判断宫颈癌患者是否存在淋巴结扩散方面展现出巨大潜力。Objective To evaluate the application value of MRI radiomics parameters in predicting lymph node metastasis in cervical cancer patients,and to establish a model based on clinical information,in order to achieve effective prediction of lymph node metastasis status of cervical cancer before surgery.Methods A retrospective analysis was performed for the data of 171 patients with cervical cancer who underwent pelvic MRI examination and confirmed by surgical pathology in the First Hospital of Shanxi Medical University from October 2018 to September 2023,and these patients had both lymph node metastasis and no metastasis.The training set(n=120)and the test set(n=51)were randomly assigned according to the ratio of 7:3.For each participant,the region of interest(ROI)of the lesion was manually delineated by multi-sequence MRI images,and the feature values were extracted from it using a Python program.The combination of Pearson correlation coefficient and LASSO regression was used to screen the features and reduce the dimensionality.SPSS 26.0 software was used to complete the statistical analysis of continuous variables and categorical data.MedCalc software was used to generate receiver operating characteristic(ROC)curves to evaluate the ability of different models to identify lymph node metastases in cervical cancer.Results The results showed that the area under the curve(AUC)of the model based on clinical information alone,the pure radiomics model and the comprehensive model combining the advantages of the two were 0.759,0.828 and 0.848 in the training set,respectively,while in the independent validation cohort,these three values were 0.809,0.806 and 0.848,respectively,indicating that the comprehensive model had the best performance.In addition,by plotting the nomogram and its calibration curve,a good agreement between the predicted probability and the actual observation can be observed.The decision curve analysis further confirmed the significant clinical utility of this comprehensive strategy.Conclusion The new model,w

关 键 词：宫颈癌 淋巴结转移 磁共振成像 影像组学 

分 类 号：R711.74[医药卫生—妇产科学]