Behroozi Mehdi, Daliri Mohammad Reza
Biomedical Engineering Department, Faculty of Electrical Engineering, Iran University of Science and Technology (IUST), Narmak, Tehran 16846-13114, Iran , School of Cognitive Sciences (SCS), Institute for Research in Fundamental Science (IPM), P. O. Box 19395-5746, Niavaran, Tehran, Iran.
J Integr Neurosci. 2014 Dec;13(4):645-67. doi: 10.1142/S0219635214500241.
Recently, the multivariate analysis methods have been widely used for predicting the human cognitive states from fMRI data. Here, we explore the possibility of predicting the human cognitive states using a pattern of brain activities associated with thinking about concrete objects. The fMRI signals in conjunction with pattern recognition methods were used for the analysis of cognitive functions associated with viewing of 60 object pictures named by the words in 12 categories. The important step in Multi Voxel Pattern Analysis (MVPA) is feature extraction and feature selection parts. In this study, the new feature selection method (accuracy method) was developed for multi-class fMRI dataset to select the informative voxels corresponding to the objects category from the whole brain voxels. Here the result of three multivariate classifiers namely, Naïve Bayes, K-nearest neighbor and support vector machine, were compared for predicting the category of presented objects from activation BOLD patterns in human whole brain. We investigated whether the multivariate classifiers are capable to find the associated regions of the brain with the visual presentation of categories of various objects. Overall Naïve Bayes classifier perfumed best and it was the best method for extracting features from the whole brain data. In addition, the results of this study indicate that thinking about different semantic categories of objects have an effect on different spatial patterns of neural activation, and so it is possible to identify the category of the objects based on the patterns of neural activation recorded during representation of object line drawing from participants with high accuracy. Finally we demonstrated that the selected brain regions that were informative for object categorization were similar across subjects and this distribution of selected voxels on the cortex may neutrally represent the various object's category properties.
最近，多变量分析方法已被广泛用于从功能磁共振成像（fMRI）数据预测人类认知状态。在此，我们探索利用与思考具体物体相关的大脑活动模式来预测人类认知状态的可能性。功能磁共振成像信号结合模式识别方法，用于分析与观看12个类别中由单词命名的60张物体图片相关的认知功能。多体素模式分析（MVPA）中的重要步骤是特征提取和特征选择部分。在本研究中，针对多类别功能磁共振成像数据集开发了新的特征选择方法（准确率方法），以便从全脑体素中选择与物体类别对应的信息性体素。这里比较了三种多变量分类器的结果，即朴素贝叶斯、K近邻和支持向量机，用于根据人类全脑激活的血氧水平依赖（BOLD）模式预测所呈现物体的类别。我们研究了多变量分类器是否能够找到与各种物体类别的视觉呈现相关的大脑区域。总体而言，朴素贝叶斯分类器表现最佳，是从全脑数据中提取特征的最佳方法。此外，本研究结果表明，思考不同语义类别的物体对神经激活的不同空间模式有影响，因此有可能根据参与者在物体线条图呈现过程中记录的神经激活模式高精度地识别物体类别。最后，我们证明了对物体分类有信息价值的所选脑区在不同受试者之间是相似的，并且这些所选体素在皮层上的分布可能中性地代表各种物体的类别属性。
