【作者】 李军；

【导师】 石光明； 田捷；

【作者基本信息】 西安电子科技大学 ， 电路与系统， 2010， 博士

【摘要】 面孔是人类社会生活中非常重要的一类视觉刺激物,人与人之间的交流方式在很大程度上依赖于面孔信息的加工,并且人们拥有高超的面孔加工技能。面孔加工如何在大脑中实现一直是多个学科领域的研究热点,随着研究的深入,在这个问题上取得了不少成果,但是其神经机制仍然很不清楚。本文的目的是利用fMRI成像技术来研究面孔认知加工中的若干问题,希望能够推动面孔加工研究的进展。近年来,采用多种成像技术对面孔加工脑激活模态的研究都有一致的发现:相对于其他视觉刺激物,面孔总能引起梭状回中部区域很强的激活,并定义该区域为梭状回面孔区(FFA)。目前对该区域功能的解释存在两种对立理论:一种理论认为该区域就为面孔特有的加工区域;而另外一种解释理论认为该区域的激活是由于人们是面孔加工的专家而导致的。在本文第一个研究中,我们分析了提出这些理论的研究中所使用的视觉刺激控制物,发现它们同面孔不匹配。为此,我们通过改进对比控制物来研究面孔加工神经区域特异性的问题。对中国人而言,汉字字符在特征属性多个维度上同面孔相似,它是面孔最理想的对比条件。因此,汉字在本文研究中被选择为面孔的对比控制物。研究结果表明,面孔和汉字引起双侧梭状回的激活模式有非常强的相似性和相关性,这个结果应该由面孔和汉字在特征属性多维度相似而导致的。但是,面孔对比汉字引发右侧梭状回很强的激活,而汉字对比面孔在双侧梭状回都没有引起激活。这个结果有力的支持了FFA区域存在对面孔加工特异性神经细胞的假设。面孔加工有两种实现方式:bottom-up和top-down方式。由于bottom-up方式下面孔加工实验易于实现和控制,故已有的面孔加工研究主要集中在该方式下。对于面孔top-down加工的研究,由于面孔top-down加工信号流的提取易受到bottom-up方式下的面孔加工信号的干扰,因而设计合理有效的实验范式来提取面孔top-down加工信号流成为首要一步。在本文第二个研究中,一个新颖的实验范式被设计出来研究面孔top-down加工神经机制:经过训练的被试从纯噪声图片中感知虚幻面孔。由于所感知的噪声图片物理属性一致,被试从中感知到的虚幻面孔完全由top-down方式产生,因而能提纯到面孔top-down加工脑激活模式。在此实验范式应用的基础上,传统Pearson时间序列相关分析,以及对其进行改进的心理生理交互作用(PPI)分析被用来获取top-down方式下的面孔加工神经网络。该面孔加工网络除了包括传统方式下得到的分布式面孔加工网络的核心系统区域外,同时也包括了用于处理视觉刺激低频信息的脑区,决策的脑区,以及用于工作记忆和注意力调节的脑区。紧接着,采用动态因果模型(DCM)分析方法来推断该网络中梭状回面孔区(FFA)、枕部面孔区(OFA)、眶额叶皮层(OFC)和顶下小叶(IPL)之间的有效连通模型,已有的研究认为这四个脑区在面孔加工或者top-down处理中起着重要作用。最优网络模型表明,OFC和IPL在top-down虚幻面孔加工中有着重要作用,它们能够规整OFA的激活模式:OFC能够为OFA提供可能面孔的低频特征信息;IPL能够施加top-down注意力到OFA,以协助OFA根据OFC提供的信息从纯噪声图片提取类似面孔的特征信息;然后,OFA在对这些特征信息进行初步加工后,提供给FFA做进一步的面孔整体相关信息的加工。面孔加工神经机制一直是多个学科领域的研究热点,也在各个研究领域中取得许多显著成果。然而,面孔加工神经机制是什么这个问题的研究亟待深入。借助多学科交叉优势,能够推动面孔感知神经机制研究的进展,能够为人类面孔加工缺陷的疾病提供治疗手段,同时也能为模式识别中面孔识别技术提供新的思路和解决方法。在本文第三个研究中,首先回顾了在面孔感知研究中有着重要影响的功能认知模型以及分布式面孔感知神经模型。然后综合当前采用各种先进技术及实验范式的研究成果,推导出面孔加工神经网络可能的时间模式和空间分布模式,希望能够对以后面孔加工神经机制进一步深入研究提供解决思路。更多还原

【Abstract】 Faces are one of the upmost important visual stimuli in the human’s social lives; the means of communication among people rely heavily on processing the information contained in the faces, and humans are exceptionally skilled at face perception. How the face processing is performed in the brain has been the research focus in numerous disciplines. With the development of neuroscience, many achievements on this question have been gained. However, it is still unclear what the neural mechanisms of face processing are. Therefore, the goal of this article is to study a number of cognitive issues of face processing based on fMRI, hoping to promote this research progress.In recent years, converging evidences, which come from the studies of activation patterns in face processing using a variety of functional imaging technologies, have provided a very consistent finding, which is that faces always lead strong activation in the fusiform gyrus when compared to other visual stimuli. This area is defined as the fusiform face area (FFA). At present, there are two distinct different explanations on the function of this region: one theory interprets that this area is just specialized for face processing, but the other one considers that this area is responsible for faces as human are the experts in face processing. In our first study, we find that the control stimuli used in the previous studies are not matched with the faces. Thus, we improve the experimental control stimuli of face to study the neural specificity of face perception. For literate Chinese adults, the characters should be the ideal contrast condition for face stimuli, because it is similar to the face in multi-dimensions. Hence, in the first study, characters are chosen as the comparison stimuli of faces. The results in this study indicate that the activation patterns of bilateral fusiform gyrus induced by faces and characters are very similar and strong correlative,which is attributed to the similar multi-dimensions between the faces and characters. However, it is found that the right fusiform gyrus is strongly activated when faces relative to characters. On the opposite, bilateral fusiform gyrus is not active when characters comparing to faces. Those results convincingly support the hypothesis that the neurons specified for face processing exist in this region.Face processing can be achieved through two means: bottom-up and top-down. For the experimental approach of bottom-up face processing is easily to be implemented and controlled, most of previous studies of face processing have focused on this means of face processing. In the study of top-down face processing, as the information coming from the bottom-up face processing easily interferes to extract the brain activation patterns of top-down face processing, it is the first key step to design a logical and effective paradigm to obtain the signal flow of top-down face processing. In the second study, a novel experimental paradigm is designed to study the neural substrate of top-down face processing. Subjects are trained to perceive illusory faces from the pure noise images. Since the physical properties of all noise images are identical and illusory face processing carried by subjects is formed entirely under the top-down approach.Based on the experimental paradigm, the traditional Pearson correlation analysis of time series and improved Psychophysiological Interactions analysis are used to obtain the distributed neural network of top-down face processing. This network not only contains the regions of core system in the distributed face processing network defined by previous studies, but also the brain regions involved in the processing of low-frequency information of visual stimuli, making decision, as well as in working memory and attention regulation. Further researches use the method of dynamic causal model (DCM) analysis to infer the effective connectivity network of the regions of the fusiform face area (FFA), the occipital face area (OFA), the inferior parietal lobule (IPL) and the orbitofrontal cortex(OFC), which are contained in the top-down face processing network. The four regions are considered to play a crucial role in the face processing or top-down processing manner. The optimal network model indicates that the OFC and IPL play crucial roles in the top-down face processing by regulating the activities of the OFA. The OFC can offer OFA the low-frequency information of faces and OFA can search the pure noise images for face-like features under directing top-down visual attention exerted by IPL. After the initial processing by OFA, the face-like featured information is transmitted to FFA for further holistic face processing.The neural substrate of face processing is always the hot field in many disciplines, and many significant results have been achieved in relative research fields. Nevertheless, this issue is still not well understood. It could take advantage of the interdisciplinary efforts to promote the progress of neural mechanisms study of face perceptions, to provide the proper treatment ways for face processing diseases, and also to provide new methods and ideas to solve the bottleneck problems of face identification in pattern recognition field. In the third study of this article, those most influential functional cognitive models and neural distributed network models of face processing are primarily presented. Then recent findings using novel methods and advanced technologies to study the neural mechanism of face processing are accumulated to investigate spatiotemporal relation among the neural regions involved in face processing, hoping to provide new ideas for our further studies of face processing after this review.更多还原