基于术前CT影像特征预测胆囊癌淋巴管浸润的研究

目的 评估术前胆囊癌CT影像特征对预测胆囊癌淋巴管浸润（LVI）的价值,构建预测胆囊癌LVI的列线图模型并评价其预测效能。方法 回顾性分析116例接受术前增强CT检查的胆囊癌病人的临床、病理及影像资料。病人按6︰4比例随机分为训练集72例,验证集44例。根据病理结果将病人分为LVI阳性组及LVI阴性组。由2位放射科医生评估训练集及验证集的CT影像特征,包括肿瘤基底浸润胆囊壁范围、影像T分期和N分期、有无胆囊结石。采用独立样本t检验、Mann-Whitney U检验及χ²检验比较2组病人中各项CT征象的差异性,并对训练集采用logistics回归分析筛选独立危险因素。建立列线图预测模型,采用受试者操作特征（ROC）曲线评估预测模型的预测效能并计算相应的曲线下面积（AUC）,绘制校准曲线对列线图预测模型进行验证。采用决策曲线评价其临床净获益。结果 训练集及验证集中,LVI阳性组的CT影像N分期、肿瘤基底浸润胆囊壁范围、T分期均显著高于LVI阴性组（均P<0.05）。多因素分析结果显示影像N分期、T分期、肿瘤基底浸润胆囊壁范围为LVI独立危险因素。建立列线图预测模型,结果显示肿瘤基底浸润范围的得分最高,其次是N分期,再次是T分期。决策曲线分析显示模型的净获益良好,训练集和验证集中CT影像特征预测LVI的ROC曲线的AUC分别为0.828（95%CI：0.721~0.907）、0.752（95%CI：0.599~0.870）,差异无统计学意义（P>0.05）。训练集AUC的截断值为0.521时,模型在训练集中的敏感度、特异度分别为0.629和0.892;验证集AUC的截断值为0.432时,对应的敏感度和特异度分别为0.778和0.654。结论 联合T分期、N分期及肿瘤基底浸润范围建立的列线图预测模型,能够有效预测胆囊癌LVI。

Preoperative assessment of lymphovascular invasion of gallbladder cancer with CT-based image features

Objective To evaluate the value of preoperative CT signs of gallbladder cancer in predicting lymphatic infiltration (LVI) of gallbladder cancer, construct nomogram model for predicting LVI of gallbladder cancer and evaluate its predictive performance. Methods Clinical, pathological and imaging data of 116 patients who had undergone preoperative enhanced CT for gallbladder cancer were retrospectively analyzed. Patients were randomly divided into training set (n=72) and validation set (n=44) at a ratio of 6∶4. Patients were divided into LVI positive group and LVI negative group according to pathological results. The CT image features of the training set and validation set were evaluated by 2 radiologists, including the tumor base invasion of gallbladder wall, image T-stage and N-stage, and the presence of gallbladder stones. The differences in each CT sign between the two groups of patients with and without LVI were analyzed using independent samples t-test, Mann-Whitney U test and χ² test, and their independent risk factors were screened by logistic regression analysis to create a nomogram of the prediction model, and receiver operating characteristic (ROC) curves were used to assess the predictive power of the nomogram prediction model and the corresponding area under the curves (AUCs) was calculated. The nomogram prediction model was validated by drawing calibration curves. The net clinical benefit of the model was assessed using decision curves. Results In the training set and validation set, N stage, T-stage and tumor base invasion of gallbladder wall in LVI positive group were significantly higher than those in LVI negative group (all P<0.05). Multivariate regression analysis showed that N-stage, T-stage, and tumor base invasion of gallbladder wall were independent risk factors for predicting LVI. The nomogram prediction model was further established, and the model showed the highest score for tumor base invasion of gallbladder wall, followed by N-stage, and then T-stage. Decision curve analysis showed that the net benefit of the prediction model was good. The AUCs of the ROC curves for predicting LVI with CT image features in the training and validation sets were 0.828 (95%CI: 0.721-0.907) and 0.752 (95%CI: 0.599-0.870), respectively, and the difference was not statistically significant (P>0.05). The sensitivity and specificity of the model in the training set were 0.629 and 0.892, respectively, when the AUC cutoff value of the training set was 0.521; the corresponding sensitivity and specificity were 0.778 and 0.654, respectively, when the AUC cutoff value of the validation set was 0.432. Conclusion The combination of T stage, N stage, and range of tumor basal infiltration of the gallbladder wall can successfully predict the LVI of gallbladder carcinoma.