基于黑质铁沉积的QSM影像组学在帕金森病诊断中的价值研究

A Study on the Value of QSM Radiomics Based on Iron Deposition in Substantia Nigra in Diagnosing of Parkinson’s Disease

目的应用影像组学纹理分析探讨定量磁敏感图(QSM)在帕金森病(PD)诊断中的价值。方法 58例PD确诊患者及28名健康对照者(HC)均进行QSM检查,将QSM数据进行后处理,然后导入软件并手动勾画感兴趣区(ROI),再对ROI进行纹理特征提取。每例受检者提取1132个纹理特征,降维后筛选出对PD诊断贡献较大的少数特征。按照7∶3的比例将参与者随机分为训练组与测试组,创建多元逻辑回归模型、计算影像组学评分(Radscore)、基于Radscore创建列线图。应用Mann-Whitney U检验或t检验,比较PD组与HC组间各参数的差异,对于差异有统计学意义的参数,绘制受试者工作特征(ROC)曲线。结果 PD组与HC组的影像组学特征及Radcore值均有差异(P<0.05);多元逻辑回归模型在训练组与测试组的曲线下面积(AUC)分别为0.90、0.84,Radcore的AUC为0.85,显示模型具有良好的分类性能;多元逻辑回归模型在训练组与测试组的阈值(0.87、0.87)、Radcore的阈值(0.06)为诊断PD提供了最佳敏感性与特异性的组合。结论 QSM影像组学纹理分析可以鉴别PD及HC,对于PD的诊断具有重要的参考价值。

Objective To explore the effect of quantitative susceptibility mapping(QSM) in diagnosing of Parkinson’s disease(PD) by using texture analysis of radiomics. Methods 58 patients with PD and 28 healthy controls(HC) underwent QSM.All the data after post-processing were imported into a software and the regions of interest(ROI) were manually delineated.Then, 1132 features of radiomics in the ROI were extracted from each participant. Feature dimensionality reduction was carried out and a small number of features which made a great contribution to diagnose PD were selected.Participants were randomly divided into training group and test group according to the ratio of 7:3. Based on these features, multiple Logistic regression model was created and radiomics score(Radscore) of each participant was calculated.A nomogram for PD risk assessment was created based on the patients’ Radscores.The differences in features between PD and HC were compared using the Mann-Whitney U test or t-test.Receiver operating characteristic(ROC) curves were implemented for the parameters with significant differences. Results There were significant differences in the features of radiomics between PD and HC,as well as the Radscore(P< 0.05);The ROC analysis revealed that the area under the curve(AUC) of the multiple Logistic regression model of the training group and test group(0.90,0.84) revealed good classification performance of the model, which is further illustrated by the AUC of Radscore(0.85).The cut-off values of multiple Logistic regression model in the training group and test group(0.87,0.87) provided the best combination of sensitivity and specificity to distinguish PD from HC,the cut-off values of Radscore(0.06) played the similar role. Conclusion QSM texture analysis of radiomics performs well in identification of PD from HC,which contributes to diagnose PD.