多参数MRI的影像组学融合模型在术前预测宫颈癌淋巴结转移的应用价值

目的探讨基于多参数MRI及临床特征的融合模型在术前预测宫颈癌患者淋巴结转移的价值。方法回顾性分析山西省肿瘤医院2016年6月—2019年3月经病理证实为宫颈鳞癌并于术前行MRI检查的168例患者的资料。按照7∶3的比例,采用完全随机法将所有患者分为训练组115例和验证组53例。由两名影像科医师在MRI图像上手动勾画三维容积感兴趣区(VOI),并进行一致性分析。根据临床手术病理结果将所有患者分为淋巴结转移阴性(LN-)和阳性(LN+),临床及影像资料也对应分组。分别基于每例患者的T2WI、表观扩散系数(ADC)和增强T1WI(cT1WI)序列图像上均提取3111个影像组学特征,然后对训练组采用以最大相关最小冗余(MRMR)和最小绝对收缩与选择(LASSO)回归为主的四步法进行特征选择和影像组学标签的构建,并进行分层分析。通过多变量逻辑回归筛选独立临床危险因素并联合影像组学标签构建影像组学融合模型,并制作列线图。采用ROC曲线、校正曲线、决策分析曲线(DCA)评估列线图的预测性能及临床效益。结果训练组和验证组患者基线资料差异均无统计学意义(P值均>0.05)。基于T2WI、ADC和cT1WI合并特征降维后共得到6个影像组学特征(P值均<0.05),其中包括3个小波类特征参数和3个LoG类特征参数,均与淋巴结转移显著相关。单序列影像组学标签在训练组中ROC曲线下面积(AUC)值为0.763和0.829,显示具有良好的预测效能,合并上述序列构建的影像组学标签对应的AUC值0.859,其诊断效能优于其中任意单一序列,并在验证组得到验证。联合影像组学标签和MRI评价淋巴结状态构建的列线图在训练组和验证组中均显示出良好的鉴别能力和校正性能,对应的AUC分别为0.865和0.861。在独立验证组中的决策曲线示,当风险阈值>10%时,采用影像组学方法预测LN+的净收益优于将所有患者都看作LN+或LN-,也优于MRI评价淋巴结状态。结论通过联合基于多参数MRI的影像组学标签和MRI评价淋巴结状态建立的融合模型可作为术前评估宫颈癌淋巴结转移的一种辅助方法。

Multiparametric magnetic resonance imaging-based radiomics model for the preoperative prediction of lymph node metastasis in cervical cancer

Objective This study aimed to investigate the value of the combined radiomics model on the basis of multiparametric MRI and clinical features to predict lymph node(LN) metastasis in patients with cervical cancer. Methods A total of 168 consecutive patients with cervical cancer in the Shanxi Province Tumor Hospital from June 2016 to March 2019 were enrolled in our retrospective study. The patients were divided into a training set of 115 cases and a validation set of 53 cases via the completely stochastic method at a ratio of 7∶3. The volume of interest was delineated manually and separately by two radiologists in MR imaging, and repeatability was assessed. LN were dichotomized in accordance with the pathological Results of the operation. The clinical and the imaging data were divided into corresponding groups. A total of 3 111 imaging features were extracted from T₂-weighted image (T₂WI), apparent diffusion coefficient (ADC), and contrast-enhanced T₁-weighted image (cT₁WI) for each patient. Four-step procedures, namely, the minimum redundancy-maximum relevance and the least absolute shrinkage and selection operator regression, were applied for the feature selection and radiomics signature building. Stratified analyses were also performed. The combined radiomics model, including the clinical risk factors and the abovementioned radiomics signature, was constructed via the multivariate logistic regression method, and the corresponding nomogram was constructed. The prediction performance was determined through its calibration, discrimination, and clinical usefulness.Results No significant difference in baseline data existed between the training and the validation groups (all P values>0.05). The radiomics signature derived from the combination of T₂WI, ADC, and cT₁W images, which was composed of six LN status-related features (i.e., three wavelet and three LoG features), was significantly associated with lymph node metastasis (all P values<0.05). The radiomic signature derived from single images yielded the area under the ROC curve (AUC) values of 0.763 and 0.829 in the training set, showing good prediction performance. The radiomics signature from the aforementioned sets yielded the highest AUC (0.859), thereby showing better prediction performance than that of the signatures derived from either of them alone in both sets, as validated in the validation cohort. The radiomics nomogram that incorporated the radiomics signature and the MRI-reported LN status, also showed good calibration and discrimination in both sets, with AUCs of 0.865 and 0.861, respectively. When the threshold probability was more than 10%, the use of the radiomics nomogram to predict LN metastasis provided a better net benefit than those of the scheme in which all patients are LN-positive + or the scheme in which all patients were LN-negative and the MRI-reported LN status. The decision curve analysis confirmed its clinical usefulness.Conclusions The proposed MRI-based radiomics nomogram has good performance in predicting lymph node metastasis and may be useful in supplementing morphological evaluation to determine LN status in patients with cervical cancer.