Abstract: Feature selection and classification is the core issue for high-dimensional functional Magnetic Resonance Imaing（fMRI） data analysis. Feature selection is divided into two steps. Region of interesting is selected in brain in the first place, and the voxels of the most distinguishing stimulating task are selected among them. The method is simple, stable and consistent with human logic. Gaussian Naive Bayes（GNB） and Support Vector Machine（SVM） classifier is used to evaluate the method. The experimental results show that the method is feasible, and to compare classification methods, SVM is superior overall to GNB.

Key words: Gaussian Naive Bayes, Support Vector Machine, functional Magnetic Resonance Imaging（fMRI）, feature selection

摘要： 特征选择和分类是脑功能磁共振成像（fMRI）数据分析的核心问题。针对fMRI高维数据，特征选择分两步，选取感兴趣脑区，选择最能区分刺激任务体素。该方法简单，稳定，符合人的思维逻辑。分类器选择高斯朴素贝叶斯（GNB）和支持向量机（SVM），评估该特征选择方法。实验结果表明，该方法有效提高了分类速度，分类准确度也得到很大提高。对分类方法进行比较，SVM总体上优于GNB。

关键词: 高斯朴素贝叶斯, 支持向量机, 功能磁共振成像, 特征选择