弥散峰度成像影像组学评估帕金森病

Diffusion kurtosis imaging radiomics for evaluating Parkinson disease

目的观察弥散峰度成像(DKI)影像组学评估帕金森病(PD)的价值。方法回顾性分析76例PD患者(PD组)及80名健康对照(HC)者(HC组),按8∶2比例将其分为训练集(n=125,PD 61例、HC 64名)与测试集(n=31,PD 15例、HC 16名)。于颅脑DKI平均峰度(MK)图中自动勾画双侧黑质、尾状核、壳核、苍白球及丘脑ROI,获取其MK平均值(MK_(mean))、提取其纹理特征并进行组间比较。基于50个最优纹理特征构建支持向量机(SVM)模型,绘制受试者工作特征(ROC)曲线,计算曲线下面积(AUC),观察其评估PD的效能。结果PD组双侧黑质、尾状核及丘脑MK mean均低于HC组(P均<0.05);组间双侧壳核及苍白球MK_(mean)差异均无统计学意义(P均>0.05)。SVM模型评估训练集PD的敏感度、特异度、准确率及AUC分别为86.89%、93.75%、90.40%及0.982,在测试集分别为86.67%、93.75%、90.32%及0.958。结论DKI影像组学可通过反映脑核团微结构改变而有效评估PD。

Objective To observe the value of diffusion kurtosis imaging(DKI)radiomics for evaluating Parkinson disease(PD).Methods Totally 76 PD patients(PD group)and 80 healthy controls(HC group)were retrospectively analyzed.The subjects were divided into training set(n=125,including 61 PD and 64 HC)and test set(n=31,including 15 PD and 16 HC)at the ratio of 8∶2.ROI of bilateral substantia nigra,caudate nucleus,putamen,globus pallidus and thalamus were automatically delineated on mean kurtosis(MK)images of cerebral DKI.The mean MK_(values)(MK_(mean))of the above ROIs were obtained and compared between groups.Support vector machine(SVM)model was constructed based on 50 selected optimal texture features.Receiver operating characteristic(ROC)curve was drawn,and the area under the curve(AUC)was calculated to evaluate the efficacy of SVM model for evaluating PD.Results MK mean of bilateral substantia nigra,caudate nucleus and thalamus in PD group were all significantly lower than those in HC group(all P<0.05).No significant difference of MK mean of bilateral putamen nor globus pallidus was found between groups(all P>0.05).The sensitivity,specificity,accuracy and AUC of SVM model for evaluating PD in training set was 86.89%,93.75%,90.40%and 0.982,respectively,which in test set was 86.67%,93.75%,90.32%and 0.958,respectively.Conclusion DKI radiomics could be used to effectively evaluate PD through description of microstructural changes of cerebral nuclei.