基于二维超声和实时组织弹性成像的双模态影像组学诊断高尿酸血症患者并发痛风性关节炎的应用价值

Application value of bimodal radiomics based on two-dimensional ultrasound and real-time tissue elastography in the diagnosis of hyperuricemia patients complicated with gouty arthritis

目的探讨基于二维超声和实时组织弹性成像(RTE)的双模态影像组学在诊断高尿酸血症(HUA)患者并发痛风性关节炎(GA)中的应用价值。方法选取于我院就诊的HUA患者59例,其中并发GA患者41例,未并发GA患者18例,使用最小绝对收缩和选择算子(LASSO)回归5折交叉验证法分别对二维超声、RTE的影像组学特征和临床特征进行筛选,获取影像组学最优特征子集;然后使用支持向量机(SVM)分类器进行GA与非GA二分类诊断,绘制受试者工作特征(ROC)曲线分析Adaboost融合SVM分类器对HUA并发GA的诊断效能。结果基于二维超声图像筛选出14个非零系数特征,基于RTE图像筛选出18个非零系数特征,将2种类型特征串联融合筛选出16个非零系数特征。双模态分类效果的最佳精确度(ACC)、全组F1值分别为74.17%±3.72%、74.56%±5.22%,较二维超声模态的最佳ACC、全组F1值分别提高了3.88%、3.26%,较RTE模态的最佳ACC、全组F1值分别提高了6.61%、8.08%。Adaboost融合SVM分类器后模型的最佳ACC、全组F1值分别为76.24%±2.50%、75.73%±3.22%,较SVM模型分别提高了2.07%、1.17%。ROC曲线分析显示,0、1类作为正样本时的曲线下面积分别为0.718、0.910。结论基于二维超声和RTE的双模态影像组学特征可对HUA患者并发GA进行定量表征和有效预测,在GA早期诊断中具有潜在的临床应用价值;且二维超声的分类效果优于RTE,双模态分类效果优于单模态。

Objective To investigate the application value of bimodal radiomics based on two-dimensional ultrasound and real-time tissue elastography(RTE)in the diagnosis of hyperuricemia(HUA)patients complicated with gouty arthritis(GA).Methods Fifty-nine patients with HUA treated in our hospital were selected,including 41 patients complicated with GA and 18 patients without GA.LASSO regression 5-fold cross-validation method was used to screen the imaging features of two-dimensional ultrasound,RTE and clinical features,and the optimal feature subset of image omics was obtained.Support vector machine(SVM)was used for GA and non-GA binary classification diagnosis,and receiver operating characteristic(ROC)curve was drawn to evaluate the diagnostic efficiency of Adaboost fusion SVM classifier for HUA complicated with GA.Results Totally 14 non-zero coefficient features were selected based on two-dimensional ultrasonic images,18 non-zero coefficient features were selected based on RTE images,and 16 non-zero coefficient features were selected by series fusion of two types of features.The optimal most ACC and F1 values were 74.17%±3.72% and 74.56%±5.22%,respectively,which were increased by 3.88% and 3.26% compared with the most ACC and F1 values by two-dimensional ultrasound alone,and by 6.61% and 8.08% compared with RTE alone.The optimal most ACC and F1 values of the Adaboost algorithm combined with SVM classifier were 76.24%±2.50% and 75.73%±3.22%,respectively,which were increased by 2.07% and 1.17% compared with SVM primary classification model.ROC curve analysis showed that the area under the curve of class 0 and class 1 as positive samples were 0.718 and 0.910,respectively.Conclusion The bimodal radiomics features based on two-dimensional ultrasound and RTE can quantitatively characterize and effectively predict HUA patients complicated with GA,which has potential clinical application value in the early diagnosis of GA.Moreover,the classification effect of the single two-dimensional ultrasound mode is better than that of the single RTE mode,and the classification effect of the dual mode is better than that of the single mode.