中国癌症杂志 ›› 2021, Vol. 31 ›› Issue (6): 460-467.doi: 10.19401/j.cnki.1007-3639.2021.06.004

• 论著 • 上一篇    下一篇

基于多序列MRI与多体系影像组学模型预测子宫颈癌淋巴结转移的研究

董诗洁,胡晓欣,王 葳,杨 孟,岳 磊,童 彤,顾雅佳   

  1. 复旦大学附属肿瘤医院放射诊断科,复旦大学上海医学院肿瘤学系,上海 200032
  • 出版日期:2021-06-30 发布日期:2021-07-09
  • 通信作者: 胡晓欣 E-mail: 1huxx@163.com
  • 基金资助:
    上海市卫生和计划生育委员会面上项目(201940288)。

Prediction of lymph node metastasis of cervical cancer based on multi-sequence MRI and multi-system imaging omics model

DONG Shijie, HU Xiaoxin, WANG Wei, YANG Meng, YUE Lei, TONG Tong, GU Yajia   

  1. Department of Radiology, Fudan University Shanghai Cancer Center, Department of Oncology, Shanghai Medical College, Fudan University, Shanghai 200032, China
  • Published:2021-06-30 Online:2021-07-09
  • Contact: HU Xiaoxin E-mail: 1huxx@163.com

摘要: 背景与目的:术前寻找可早期用于准确评估淋巴结转移与否的生物标志物极具临床应用价值。探讨MRI影像组学参数预测子宫颈癌淋巴结转移的价值,建立和验证用于术前预测子宫颈癌淋巴结转移的影像组学模型。方法:回顾性分析2015年6月—2019年9月在复旦大学附属肿瘤医院经术后病理学检查证实的子宫颈癌非淋巴结转移患者和子宫颈癌淋巴结转移患者共202例的临床资料,所有患者均经过术前MRI检查。选用MRI图像分别为T2加权图像(T2 weighted image,T2WI)和T1增强图像(T1 contrast +,T1C+)。使用ITK-SNAP软件进行三维手动分割子宫颈癌肿瘤区域。通过开源的python包Pyradiomics和python编程平台jupyter notebook,经过10种图像类型体系和6种特征体系来提取影像组学特征,选取子宫颈癌患者202例,其中未发生淋巴结转移的104例,发生淋巴结转移的98例。T2WI序列和T1C+序列模型分别提取1 923个特征,T2WI联合T1C+序列提取3 846个特征。通过建立影像组学标签,经过机器学习模型验证影像组学标签。最后将训练集和测试集的曲线下面积(area under curve,AUC)、准确率、阳性预测值(positive predictive value,PPV)和阴性预测值(negative predictive value,NPV)作为评估影像组学标签的定量表现。结果:T2WI序列选取特征排序前14名的特征进行分类器训练,训练集AUC=0.810,测试集AUC=0.773。对于T1C+序列选取了特征排序前16名的特征进行分类器训练,训练集AUC=0.819,测试集AUC=0.781。在T2WI联合T1C+序列中选取了特征排序前16名的特征进行分类器训练,训练集AUC=0.841,测试集AUC=0.803。结论:T2WI联合T1C+序列影像组学模型对早期子宫颈癌淋巴结转移有较好的预测能力。

关键词: 子宫颈癌, 淋巴结转移, 磁共振成像, 影像组学

Abstract: Background and purpose: It is of great clinical value to search for early biomarkers that can be used to accurately evaluate lymph node metastasis before surgery. This study aimed to investigate the value of magnetic resonance imaging (MRI) omics parameters in predicting cervical cancer lymph node metastasis, and to establish and verify an imaging omics model for preoperative prediction of cervical cancer lymph node metastasis. Methods: The clinical data of 202 patients with non lymph node metastasis and lymph node metastasis of cervical cancer confirmed by postoperative pathological examination in Fudan University Shanghai Cancer Center from June 2015 to September 2019 were retrospectively analyzed. MRI images were selected as T2 weighted images (T2WI) and T1 contrast + (T1C+). Itk-snap software was used for three-dimensional manual segmentation of cervical cancer tumor regions. Through Pyradiomics, an open source Python package, and a Python programming platform Jupyter, imaging omics features were extracted through ten image type systems and 6 feature systems. Among a total of 202 patients with cervical cancer, 104 had no lymph node metastasis, and 98 had lymph node metastasis. Imaging features were extracted from each patient in each group, including 1 923 features from the lymph node metastasis group and no lymph node metastasis group of T2WI sequence, 1 923 features from the lymph node metastasis group and no lymph node metastasis group of T1C+ sequence, and 3 846 features from the lymph node metastasis group and no lymph node metastasis group of T2WI combined with T2WI-T1C+ sequence. Imaging omics label was established and validated by machine learning model. Finally, area under curve (AUC), accuracy, positive predictive value (PPV) and negative predictive value (NPV) of the training set and the test set were used as the quantitative performance of the imaging omics label. Results: The T2WI sequence selected the features in the first 14 for classifier training, with the AUC of the training set=0.810 and the AUC of the test set =0.773. For T1C+ sequence, the first 16 features of feature sequencing were selected for classifier training, with AUC=0.819 in the training set and AUC=0.781 in the test set. In T2WI combined with T1C+sequence, the first 16 features of feature sequencing were selected for classifier training, with AUC=0.841 in the training set and AUC=0.803 in the test set. Conclusion: T2WI combined with T1C+ sequential imaging omics model has a good efficacy in predicting lymph node metastasis of early cervical cancer.

Key words: Cervical cancer, Lym ph node metastasis, Magnetic resonance imaging, Radiomics