基于多序列MRI影像组学预测早期宫颈癌淋巴血管侵犯的研究

  目的  探索基于多序列MRI图像的影像组学方法在预测早期宫颈癌淋巴血管侵犯(LVSI)中的临床价值。  方法  选取2015年1月—2020年2月宁波大学附属人民医院收治经病理证实的早期宫颈癌患者134例，分层抽样选取训练组91例，验证组43例，术前均行MRI平扫、对比增强(CE-MRI)及弥散成像(DWI)检查。在T₂WI-FS、CE-MRI及DWI序列图像上分别勾画肿瘤感兴趣区，采用LASSO回归及诺模图法提取影像组学特征并建立预测模型，训练组进行特征选择分类及模型建立，验证组对构建的预测模型进行验证，分析基于MRI各序列影像组学模型对早期宫颈癌LVSI的预测效能。  结果  采用LASSO回归在早期宫颈癌患者的T₂WI-FS、CE-MRI和DWI序列影像中分别提取具有预测意义的宫颈癌LVSI的影像组学特征，WavEnLH_s-4、Horzl_LngREmph在各序列中均被筛选出。通过logistics回归模型分别构建的不同序列MRI的影像组学模型对早期宫颈癌LVSI诊断效能均较高，T₂WI-FS、CE-MRI及DWI在训练组的AUC分别为0.810、0.803和0.781，在验证组的AUC分别为0.785、0.761和0.752。使用诺模图法构建的多序列MRI影像组学在训练组的AUC为0.893，在验证组的AUC为0.859。  结论  通过诺模图法构建的基于T₂WI-FS、CE-MRI及DWI序列影像组学模型作为一种客观的影像分析方法，对早期宫颈癌LVSI具有较高的预测效能并具有一定临床应用价值。 

Multi-sequence MRI-based radiomics predicting lymph-vascular space invasion in early-stage cervical cancer

  Objective  To explore the clinical value of multi-sequence MRI-based radiomics in predicting lymph-vascular space invasion (LVSI) in early-stage cervical cancer.  Methods  A total of 134 patients (91 cases in the training group and 43 cases in the validation group) with pathological early-stage cervical cancer in the Affiliated People's Hospital of Ningbo University from January 2015 to February 2020, were retrospectively collected. All patients underwent MRI plain scan, contrast-enhanced MRI (CE-MRI) and diffusion-weighted imaging (DWI) before surgery. MRI images of each sequence were obtained, and the region of interest was drawn. Radiomic features were extracted by using the least absolute shrinkage and selection operator (LASSO) and nomogram method to construct the predictive model. The training group was used to extract feature, establish signature and construct the predictive model. The validation group was used to verify the predictive model. Receiver operating characteristic curve was used to analyse the predictive effect of each sequence MRI-based radiomic model on LVSI in early-stage cervical cancer.  Results  The predictive LVSI features were extracted from T₂WI-FS, CE-MRI and DWI sequence images in patients with early-stage cervical cancer by LASSO regression. WavEnLH_s-4 and Horzl_LngREmph were all screened out. Results showed that the diagnostic efficiency of multi-sequence MRI imaging models constructed by the Logistic regression model was high for LVSI in early-stage cervical cancer. AUC of T₂WI-FS, CE-MRI and DWI was 0.810, 0.803 and 0.781, respectively, in the training group and 0.785, 0.761 and 0.752, respectively, in the verification group. The AUC of multi-sequence MRI-based radiomic model constructed by nomogram was 0.893 in the training group and 0.859 in the verification group.  Conclusion  As an objective image analysis method, the nomographic model based on T₂WI-FS, CE-MRI and DWI sequence has a high predictive effect and certain clinical application value in LVSI of early-stage cervical cancer.