基于磁共振肝脾影像组学特征结合临床影响因素联合模型诊断显著性肝纤维化的效果评价

目的：探讨基于多模态平扫腹部MRI提取肝脏-脾脏联合影像组学特征结合临床影响因素联合模型在诊断显著性肝纤维化中的效果。方法：收集2017 年5月至2022 年5月于温州医科大学附属第二医院行经肝脏组织穿刺活检或手术病理检查证实为肝纤维化，并在病理检查6个月内接受过标准腹部MRI平扫检查的患者110例，将所有患者以7:3随机分为训练集和测试集。按照METAVIR评分系统，将F2级及以上定义为显著性肝纤维化组（62例），F2级以下定义为无或非显著性肝纤维化组（48例）。分别标注肝脏、脾脏特征，并从中提取影像组学特征经筛选后分别构建肝脏、肝脏-脾脏联合特征的支持向量机影像组学模型和影像组学标签，以此计算每位患者的影像组学评分（Rad-score）。采用Logistic回归分析显著性肝纤维化的临床影响因素。最后使用Logistic回归构建基于临床影响因素和Rad-score的联合模型，绘制列线图。使用受试者工作特征（ROC）曲线和决策曲线分析（DCA）评估模型的性能。结果：有22、36 个影像组学特征经筛选后分别用于构建肝脏、肝脏-脾脏影像组学模型。多因素Logistic回归分析结果显示，性别女性（OR =0.126，95%CI =0.040～0.354，P ＜0.001）、年龄（OR =0.985，95%CI =0.066～0.999，P =0.011）、乙肝感染（OR =5.139，95%CI =1.898～15.137，P =0.002）、APRI指数≥1（OR =3.793，95%CI =1.231～14.5，P =0.033）是独立临床影响因素，被纳入构建临床预测模型。在Logistic回归模型中，肝脏特征、肝脏-脾脏联合特征所构建的影像组学模型在ROC曲线下面积（AUC）分别为0.828和0.917，表明肝脏-脾脏联合特征影像组学模型诊断效能更优。将肝脏-脾脏联合特征影像组学模型作为影像组学预测模型与临床预测模型结合获得联合预测模型，其在训练集、测试集的AUC分别为0.948和0.963。DCA显示，联合预测模型的临床实用性最佳。结论：基于多模态平扫腹部MRI提取肝脏-脾脏联合影像组学特征较单一肝脏特征在诊断显著性肝纤维化中有更好的诊断效能，联合预测模型相比临床预测模型能进一步提高诊断效能。

Diagnosis of significant liver fibrosis: An evaluation of the combined model of extracting liver-spleen radiomics features with MRI and clinical risk factors

Objective: To investigate the effect of combining liver-spleen radiomics features based on multimodality abdominal MRI with clinical risk factors in the diagnosis of significant liver fibrosis. Methods:A total of 110 patients who underwent liver biopsy or surgical pathological examination and received standard abdominal MRI within 6 months of pathological examination were collected from the Second Affiliated Hospital of Wenzhou Medical University from May 2017 to May 2022. All patients were randomly divided into training set and testing set by 7:3. According to the METAVIR scoring system, grade F2 and above was defined as significant liver fibrosis, and grade below F2 as no or non-significant liver fibrosis. The liver and spleen features were marked respectively, and the radiomics features were extracted respectively. After feature screening, SVM machine learning radiomics models of the liver and liver -spleen combined features were constructed to calculatethe radiomics score of each patient (Rad-Score). Logistic regression was used to analyze the clinical influence factors for significant liver fibrosis. Finally, Logistic regression was used to construct a joint model based on clinical influence factors and Rad-score, and to draw a nomogram. The performance of the model was evaluated by using the receiver operating characteristic (ROC) curve and the decision curve analysis (DCA). Results: After screening, 22 and 36 radiomics features were involved in the construction of liver and liver-spleen radiomics model respectively. In the multifactorial regression analysis, results showed that gender (female) (OR=0.126,95%CI=0.040-0.354, P<0.001), age (OR=0.985, 95%CI=0.066-0.999, P=0.011), hepatitis B infection (OR=5.139,95%CI=1.898-15.137, P=0.002), and APRI index≥1 (OR=3.793, 95 %CI=1.231-14.5, P=0.033) were independent clinical influence factors, which were included in the construction of the clinical predictive models. In the Logistic regression model, the area under the ROC curve (AUC) of the liver features and liver-spleen combined feature radiomics model were 0.828 and 0.917, respectively, indicating that the diagnostic efficiency of the liver-spleen combined feature radiomics model was better. The radiomics model was combined with the clinical predictive model to establish a combined predictive model, of which AUC in the training sets and in the testing sets was 0.948 and 0.963 respectively. DCA showed that the clinical applicability of the combined predictive model was optimal.Conclusion: The combined feature radiomics model based on multimodality abdominal MRI has better diagnostic efficacy than the single liver features. And the combined predictive model has better diagnostic efficacy than the clinical predictive model.