基于扩散张量成像的纹理分析对帕金森病诊断价值的研究

Diagnostic value of texture analysis based on diffusion tensor imaging in Parkinson's disease

目的探讨基于扩散张量成像(diffusion tensor imaging,DTI)上灰质核团和脑白质的纹理分析对帕金森病(Parkinson's disease,PD)的诊断价值及其与病情发展的相关性。材料与方法前瞻性分析30例PD患者与22例正常对照者进行DTI扫描,工作站后处理得到两组的各向异性指数(fractional anisotropy,FA)图,在ITK-SNAP软件上勾画ROI,包括双侧尾状核头、苍白球、壳核、黑质、红核、小脑齿状核和双侧半卵圆区,用A.K软件进行纹理特征的提取,通过Mann-Whitney U检验、单因素Logistic回归分析、最大相关最小冗余(maximum relevance minimum redundancy,mRMR)选择5个联合诊断效能最高的纹理特征,并构建随机森林(random forest,RF)模型建立预测模型,绘制ROC曲线分析评价模型的诊断效能,并用交叉验证的方法对模型的可靠性进行评估。降维所得的纹理特征与PD患者的简易精神状态评价量表(Mini Mental State Examination,MMSE)、统一帕金森病评定量表(Unified Parkinson's Disease Rating Scale,UPDRS)及病程进行Pearson相关性分析,与Hoehn-Yahr(H-Y)分级进行Spearman相关性分析。结果降维选择后得到的5个纹理特征,ROC分析得出其独立预测PD的曲线下面积(AUC)范围为0.692~0.871,构建PD预测模型,该模型的AUC、准确度、敏感度、特异度分别为0.92、0.86、0.89、0.84。交叉验证的准确度、敏感度、特异度分别为0.89、0.84、0.94。5个纹理特征与上述各临床评分量表及病程均无显著相关性。结论基于DTI的纹理分析对PD有很高的诊断价值,但对病情发展的评估价值不大。

Objective:To investigate the diagnostic value of texture analysis of gray matter nuclei and white matter on diffusion tensor imaging(DTI)in Parkinson's disease(PD)and its correlation with the development of PD.Meterials and Methods:Thirty PD patients and 22 normal controls were prospectively collected for DTI scanning.The fractional anisotropy(FA)diagrams of the two groups were obtained by post-processing.The regions of interest(ROI),including bilateral caudate head,globus pallidus,putamen,substantia nigra,red nucleus,dentate nucleus and centrum semiovale,were delineated by ITK-SNAP software.The texture features were extracted by A.K software.The Mann Whitney U test,Univariate logistic regression analysis,mRMR(maximum relevance minimum redundancy)was applied to select 5 texture features with the highest joint diagnostic efficiency,and random forest(RF)was constructed.Receiver operating characteristic(ROC)curve analysis was used to evaluate the diagnostic efficiency of the model;besides,the cross-validation method was employed to verify the reliability of the model.In addition.The texture features obtained by dimensionality reduction were analyzed by Pearson correlation with mini mental state examination(MMSE),unified Parkinson's Disease Rating Scale(UPDRS)and course of disease,and Spearman correlation with Hoehn-Yahr(H-Y)stages.Results:Five texture features were obtained after dimensionality reduction,and AUC(area under the curve)of independent prediction of Parkinson's disease was ranging from 0.692 to 0.871 by ROC analysis.The AUC,accuracy,sensitivity and specificity of the Parkinson's disease prediction model were 0.92,0.86,0.89,0.84,respectively.The accuracy,sensitivity and specificity of cross-validation were 0.89,0.84,0.94,respectively.No significant correlation was found between the five texture features and the clinical scale of disease.Conclusions:Texture analysis based on DTI has a high diagnostic value for PD.However,the value for evaluating the disease development is limited.