摘要
帕金森病(Parkinson disease,PD)是常见的神经退行性疾病。其主要病理改变是黑质纹状体细胞的进行性损失和细胞内路易斯小体的聚集,只有到神经元死亡50%-60%后才会出现临床症状。因此PD的早期诊断一直是一个难题。该研究是基于matlab的特征数据提取、特征筛选以及应用支持向量机的分类器训练与测试。参考了ALL大脑分区模版进行特征数据提取,提取的特征包括脑脊液CSF、灰质GM、白质WM、低频振幅ALFF、区域一致性Re Ho。特征选择采用三种方法,包括:双样本T检验、基于距离的Relief排序和基于支持向量机的递归特征消除(SVM-RFE)。将特征选择得到的特征向量集用于分类器训练,分类器测试使用留一交叉验证。结果表明,基于MRI影像学分析方法,可以对早期PD的出现进行准确率较高的预测。
Parkinson＇s disease （PD） is a common neurodegenerative disease. The pathologic change of PD is the progressive damageof nigrostriatal cells and the clinical symptoms can hardly be observed until the death rate of cells rise to 50%-60%. This thesisfocuses on the feature extraction and selection, as well as construction and evaluation of classifier using Support Vector Machine（SVM）. Feature data are obtained according to Anatomical Automatic Labeling （AAL） atlas, including cerebrospinal fluid （CSF）, greymatter （GM）, white matter （WM） obtained from structure imaging and amplitude of low-frequency fluctuation （ALFF）, regionalhomogeneity （ReHo） obtained from functional imaging. Three methods are used in feature selection including Two Sample T-test,Relief algorithm based on distance sorting and Support Vector Machine Recursive Feature Elimination （SVM-RFE）. The classifier isdeveloped based on SVM. Leave One Out Cross Validation （LOOCV） method is applied to test the classifier. Result shows that it isreliable to predict the early presence of PD based on imaging analyses of MRI.
出处
《中国数字医学》
2016年第7期8-10,30,共4页
China Digital Medicine
基金
国家"863计划"项目(编号:2013AA041201
2015AA020109)
中央高校基本科研业务费专项资金~~
关键词
帕金森病
递归特征消除
支持向量机
Parkinson＇s disease, recursive feature elimination, support vector machine
