钆塞酸二钠增强MRI影像组学模型与超声弹性成像对非显著肝纤维化诊断的比较

Comparison of a Gd-EOB-DTPA-enhanced MRI radiomics model and ultrasound elastography in diagnosing non-significant liver fibrosisy

目的 比较钆塞酸二钠(Gd-EOB-DTPA)增强MRI肝胆期影像组学模型与超声弹性成像(UE)评估非显著肝纤维化的能力,联合影像组学特征等多参数构建联合模型后观察评估能力是否进一步提升。方法 回顾性纳入来自2家医院同时行Gd-EOB-DTPA增强MRI和超声弹性成像检查的慢性肝病病人201例,根据病理结果中肝纤维化程度将病人分为非显著组和显著组。按医院来源将病人分为训练集152例(非显著组38例和显著组114例)和测试集49例(非显著组11例和显著组38例)。选取2组间差异有统计学意义的临床指标进行logistic回归分析,纳入临床独立诊断因素建立临床模型。通过UE获取肝脏硬度测量(LSM)值。选择MRI肝胆期影像,提取全肝影像组学特征,采用10折交叉验证的最小绝对收缩和选择算子(LASSO)回归算法筛选特征并计算影像组学评分,同时建立影像组学模型。构建临床-超声联合模型及临床-超声-影像组学联合列线图模型。采用受试者操作特征(ROC)曲线先后分析多个模型及参数的效能,并采用DeLong检验比较效能的差异。采用校准曲线评估预测效果和实际病理分级结果的一致性。采用决策曲线分析(DCA)评估各模型的临床价值。结果 训练集和测试集中,影像组学模型与LSM、临床模型在对非显著肝纤维化的诊断效能相近,差异无统计学意义(均P>0.05)。临床-超声联合模型的诊断效能高于单独的临床模型和LSM(均P<0.05);临床-超声-影像组学联合列线图模型诊断非显著肝纤维化的效能优于其他模型和LSM值(均P<0.05)。校准曲线评估显示临床-超声-影像组学联合列线图模型预测效果与实际结果一致性良好,且DCA显示联合列线图模型在10%~92%阈值概率内的净获益率最高。结论 基于Gd-EOBDTPA增强MRI肝胆期影像的影像组学模型与UE的LSM值对非显著肝纤维化的诊断效能相似,联合多参数构建联合模型后可明显提高诊断效能。

Objective To compare the diagnostic performance of a gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid(Gd-EOB-DTPA)-enhanced MRI hepatobiliary phase radiomics model and ultrasound elastography(UE)in assessing non-significant liver fibrosis.Additionally,to investigate whether combining radiomics features with other parameters into a joint model can improve diagnostic efficiency.Methods This retrospective study included 201 patients with chronic liver disease from two hospitals who underwent both Gd-EOB-DTPA-enhanced MRI and UE.Patients were classified into non-significant and significant groups based on pathological results.The dataset was divided into a training set of 152 cases(38 non-significants,114 significants)and a test set of 49 cases(11 non-significants,38 significants)according to hospital origin.Clinical indicators with significant differences between the two groups were selected for logistic regression analysis to establish a clinical model. Liver stiffness measurement (LSM) values were obtained from UE. Radiomics features were extracted from hepatobiliary phase MRI images, and a radiomics model was developed using the least absolute shrinkage and selection operator (LASSO) regression algorithm with 10-fold cross-validation. A clinical-ultrasound combined model and a clinical-ultrasound-radiomics combined nomogram model were constructed. The performance of multiple models and parameter were sequentially analyzed using receiver operating characteristic (ROC) curves, and the performance differences were compared by DeLong test. Calibration curves were used to assess the consistency between predicted outcomes and actual pathological gradings. Decision curve analysis (DCA) was used to evaluate the clinical value of each model. Results In both the training and test sets, the diagnostic performance of the radiomics model, LSM, and clinical model for non-significant liver fibrosis was similar, with no statistically significant differences (all P>0.05). The diagnostic performance of the clinical-ultrasound combined model was higher than the clinical model and LSM (all P<0.05);the diagnostic performance of the clinical-ultrasound-radiomics combined nomogram model for diagnosing non-significant liver fibrosis was superior to the other models and LSM (all P<0.05). Calibration curves assessment showed good consistency between the predictions of the clinical-ultrasound-radiomics combined nomogram model and actual results, while DCA demonstrated that the nomogram model had the highest net benefit across a threshold probability range of 10% to 92%. Conclusion The diagnostic performance of the Gd-EOB-DTPA-enhanced MRI hepatobiliary phase radiomics model is comparable to the LSM from UE for non-significant liver fibrosis. However, combining multiple parameters into a joint model significantly improves diagnostic accuracy.