Abstract: Over the past decade fMRI researchers have developed increasingly sensitive techniques for analyzing the information represented in BOLD activity. The current understanding of the goal of many fMRI studies has been achieved by extracting representation information from analyzing fMRI data in some specific region of the brain rather than just comparing the activation difference among experiment conditions. Multivariate pattern analysis is one of the methods which have recently emerged as a promising computational technique in neuroimaging studies. Recently, multivariate pattern analysis method based on fMRI technology was widely used in the field of the neuroscience, which significantly changed the related question and methods of the studies of cognition. In specific, the multivariate pattern analysis method will be deeply applied to the study of mental imagery with the aim to improve the research of the essence and its related function of mental imagery processing. In this study, we introduced the principle and the history of multivariate pattern analysis method. In detail, we showed the advantage of the multivariate pattern analysis method when compared with that of the traditional method such as the fMRI analysis based on general linear model. Then we introduced the different stages of the multivariate pattern analysis in the neuroimaging studies including the classification, identification, and reconstruction. In the aspect of the classification, the researchers focused on analyzing data from the decoding perspective, one attempts to determine how much can be learned about the sensory stimuli, cognitive state, movement and so on, where the linear classifier such as the support vector machine and linear fisher classifier was widely used. In the aspect of the stimuli identification, the researchers interested on how to understand how activity varies in different brain regions when there is concurrent variation in the world by analyzing fMRI data with the encoding model such as the general linear model. On the aspect of the image reconstruction, the correspondence relationship between the stimuli properties and the neural activity pattern was built, then the Bayesian model was applied to reconstructed the visual image of the stimuli from the neural activity pattern in brain. Taken together, we indicated the develop process and the logic of the three stages. Based on this, we introduced the dilemma of the mental imagery study, and the necessity of the application the multivariate pattern analysis method in the studies of mental imagery. Then we specifically analyzed the advantage of application of multivariate pattern analysis in mental imagery study, in specific we discussed the basis of the approach of visualization of the mental imagery from the fMRI brain activity pattern. In this article, we introduced the basic claims of the perception prediction theory in the field of mental imagery that mental imagery serves to activate most of the same brain regions involved in the visual perception to modulate the visual stimuli processing, where the lateral occipital complex region is the critical regions. This study proposed that the realization of the visualization of the mental imagery will contribute to providing the new viewpoint and method for the field of mental imagery.

Key words: mental imagery, fMRI, multivariate pattern analysis, decoding model, encoding model

摘要： 功能磁共振技术在表象研究中得到广泛应用是表象研究追求客观化精确性的必然趋势。本文介绍了功能磁共振多变量模式分析方法及其演变历程，探讨了借助该方法实现“视觉表象可视化”的理论依据与亟待解决的关键问题。分析指出“视觉表象可视化”将为表象研究提供全新的研究视角与方法途径。

关键词: 表象, fMRI, 多变量模式分析, 解码模型, 编码模型