Yargholi Elahe', Hossein-Zadeh Gholam-Ali
School of Electrical and Computer Engineering, University College of Engineering, University of TehranTehran, Iran; School of Cognitive Science, Institute for Research in Fundamental SciencesTehran, Iran.
Front Hum Neurosci. 2016 Jul 13;10:351. doi: 10.3389/fnhum.2016.00351. eCollection 2016.
We are frequently exposed to hand written digits 0-9 in today's modern life. Success in decoding-classification of hand written digits helps us understand the corresponding brain mechanisms and processes and assists seriously in designing more efficient brain-computer interfaces. However, all digits belong to the same semantic category and similarity in appearance of hand written digits makes this decoding-classification a challenging problem. In present study, for the first time, augmented naïve Bayes classifier is used for classification of functional Magnetic Resonance Imaging (fMRI) measurements to decode the hand written digits which took advantage of brain connectivity information in decoding-classification. fMRI was recorded from three healthy participants, with an age range of 25-30. Results in different brain lobes (frontal, occipital, parietal, and temporal) show that utilizing connectivity information significantly improves decoding-classification and capability of different brain lobes in decoding-classification of hand written digits were compared to each other. In addition, in each lobe the most contributing areas and brain connectivities were determined and connectivities with short distances between their endpoints were recognized to be more efficient. Moreover, data driven method was applied to investigate the similarity of brain areas in responding to stimuli and this revealed both similarly active areas and active mechanisms during this experiment. Interesting finding was that during the experiment of watching hand written digits, there were some active networks (visual, working memory, motor, and language processing), but the most relevant one to the task was language processing network according to the voxel selection.
在当今现代生活中，我们经常接触到手写数字0 - 9。成功解码和分类手写数字有助于我们理解相应的大脑机制和过程，并对设计更高效的脑机接口有很大帮助。然而，所有数字都属于同一语义类别，且手写数字外观上的相似性使得这种解码分类成为一个具有挑战性的问题。在本研究中，首次使用增强朴素贝叶斯分类器对功能磁共振成像（fMRI）测量数据进行分类，以解码手写数字，该方法在解码分类中利用了大脑连接信息。对三名年龄在25 - 30岁之间的健康参与者进行了fMRI记录。不同脑叶（额叶、枕叶、顶叶和颞叶）的结果表明，利用连接信息显著提高了解码分类效果，并且对不同脑叶在手写数字解码分类中的能力进行了相互比较。此外，在每个脑叶中确定了最有贡献的区域和大脑连接，并发现端点之间距离较短的连接效率更高。而且，应用数据驱动方法研究了大脑区域对刺激反应的相似性，这揭示了该实验过程中既存在相似的活跃区域，也存在活跃机制。有趣的发现是，在观看手写数字的实验过程中，存在一些活跃网络（视觉、工作记忆、运动和语言处理），但根据体素选择，与任务最相关的是语言处理网络。
