【摘要】 目的 探讨术前基于SMOTE算法联合临床及MRI特征构建对单结节型双表型肝细胞癌预测的动态列线图模型及其验证。方法 选取单结节型肝细胞癌（HCC）患者187例，根据免疫组化结果分为单结节型DPHCC组128例和单结节型非DPHCC组59例。比较两组临床和MRI特征。采用合成少数类过采样技术（SMOTE）平衡两组数据至1∶1。将两组差异有统计学意义的临床和MRI特征纳入多因素二元Logistic回归分析，筛选DPHCC的独立预测因素。基于多因素预测模型构建ROC曲线和动态列线图，并评价模型临床应用价值。结果 结合SMOTE算法的多因素Logistic回归分析结果显示，甲胎蛋白（AFP）(OR=2.759,95%CI:1.414～5.383,P=0.003)、肝硬化（OR=3.141,95%CI:1.562～6.316,P=0.001）、脂质成分（OR=2.838,95%CI:1.388～5.803,P=0.004）、胆管扩张（OR=6.698,95%CI:2.453～18.292,P<0.001）、淋巴结肿大（OR=4.200,95%CI:1.962～8.991,P<0.001）是预测DPHCC的独立危险因素，性别（OR=0.309,95%CI:0.118～0.811,P=0.017）则为其独立保护因素。临床及MRI特征联合诊断DPHCC的ROC曲线下面积为0.806。动态列线图将预测模型可视化，校准曲线显示模型拟合优度高，决策曲线分析、临床影响曲线明确模型具有临床实用价值。结论 借助SMOTE算法的临床及MRI特征联合诊断术前预测模型可有效区分DPHCC和非DPHCC，动态列线图为术前预测单结节型DPHCC提供了一种可视、便捷且交互性强的诊断工具。更多还原

【Abstract】 Objective To explore the clinical value of a dynamic nomogram based on clinical and MRI characteristics for preoperative prediction of single-nodule dual-phenotype hepatocellular carcinoma(DPHCC). Methods A total of 187 patients with single-nodule HCC were enrolled and divided into the DPHCC(n=128) and non-DPHCC(n=59) groups based on immunohistochemical results. Clinical and MRI characteristics were compared between the two groups. The synthetic minority oversampling technique(SMOTE) was applied to balance the dataset at a 1∶1 ratio. Statistically significant characteristics were incorporated into multivariable binary logistic regression analysis to identify independent predictors of DPHCC. A ROC curve and dynamic nomogram derived from a multivariable prediction model were constructed. The clinical application value of the model was evaluated. Results The multivariable logistic regression analysis, combined with the SMOTE algorithm, indicated that AFP(OR= 2.759, 95% CI: 1.414-5.383, P=0.003), liver cirrhosis(OR= 3.141, 95% CI: 1.562-6.316, P=0.001), intracellular fat in tumor(OR= 2.838, 95% CI: 1.388-5.803, P=0.004), bile duct dilation(OR= 6.698, 95% CI: 2.453-18.292, P<0.001), and lymph node enlargement(OR= 4.200, 95% CI: 1.962-8.991, P<0.001) were independent risk factors for predicting DPHCC, whereas, gender(OR= 0.309, 95% CI: 0.118-0.811, P=0.017) was an independent protective factor. The area under the ROC curve for the multivariable prediction model was 0.806. A dynamic nomogram was constructed to visualize the prediction model, and the calibration curve showed a high degree of model fit. The decision curve analysis and clinical impact curve confirmed the clinical utility of the model. Conclusion The preoperative prediction model based on clinical and MRI characteristics, aided by the SMOTE algorithm, effectively distinguishes DPHCC from non-DPHCC. The dynamic nomogram provides a visual, convenient, and interactive diagnostic tool for the preoperative prediction of single-nodule DPHCC.更多还原