大脑如何表征情绪：基于fMRI的多变量模式分析证据

How do different emotional states represent in human brain?——Evidence from multi-variate pattern analysis based on functional MRI

不同的情绪状态如何在大脑中表征?学术界存在两种主流观点:类别取向主张大脑按照愤怒、厌恶、快乐等类别表征情绪,每种情绪具有特异的大脑表征模式;维度取向主张大脑按照效价与唤醒度两种维度表征情绪,具体情绪由效价和唤醒度两个基本维度构建形成.然而,传统单变量激活研究很难发现情绪在大脑中的特异性表征.随着功能磁共振分析方法的发展,研究者开始采用更为敏感的多变量模式分析研究大脑如何表征情绪.本文首先介绍了传统单变量激活分析的局限以及多变量模式分析方法的优势,然后分别介绍近年来基于多变量模式分析方法对情绪表征的研究并总结多变量模式分析方法对传统情绪理论发展的意义,最后探讨情绪表征未来可能的研究方向.

When studying emotion, there has been a longstanding controversy whether emotion should be better divided into different categories or into dimensions. There are a number of behavioral and physiological studies supporting both the categorical view and the dimension view. With the development of observational and analytical instrument, affective neuroscience is emerging as a new field that attracts researchers to explore how the brain generates and represents emotion. In this review, firstly we will introduce the superiority of a method called multi-variate pattern analysis（MVPA） in studying how different emotional states are represented in human brain. Traditional univariate fMRI data analysis characterizes the relationship between mental states and each individual brain voxel. Voxels have been seen as independent from each other and are isolated from different mental states. Due to its lack of sensitivity, there is no evidence shown that different emotional states have their own representational pattern in specific brain regions. Unlike univariate method, MVPA utilizes multi-voxel pattern to decode the information of mental states and have more sensitivity than univariate analysis. Secondly, the studies using MVPA to investigate the dimension view and category view will be briefly summarized. These two theories are both partly approved, as there are some evidences which support one view and some evidences support the other. We will illustrate two research orientations separately and comment on the shortcomings of the existing research. For example, the dimension view has often been studied on one singular dimension during which the other dimension is kept constant. However, emotion is constructed of both valence and arousal. Therefore, researchers should integrate both valence and arousal to examine the dimension view in the brain. Next, we will discuss what impact does MVPA bring on traditional emotion theory. Because of its incomparable advantages, MVPA could solve some controversial issues which cannot be studied only by behavioral measures, such as questions like whether positive and negative emotions are two sides of a same dimension or they are in different dimensions. or questions like the number of basic emotional states. What is more, MVPA can directly compare different emotion theories. Last but not least, we highlight the future directions about the representation of emotion in the brain. For example, some studies have shown that emotion perception could be divided into different processing stages. Therefore, we propose that the way our brain representing emotional states is different during different time course. Future studies may utilize MVPA to decode different emotional states in the time domain by using spatiotemporal pattern similarity analysis with EEG or MEG data. In addition, it is also a good direction to investigate how different brain regions interact with each other to represent emotions by resorting to network method and multi-connection pattern analysis.