如何利用开源工具分析fMRI数据:基于多体素模式的有监督机器学习算法介绍  

How to Analyze fMRI Data with Open Source Tools:An Introduction to Supervised Machine Learning Algorithm for Multi-Voxel Patterns Analysis

作　　者：王钰莹 于海峰 黎兵[2] 鲁学明 Wang Yuying;Yu Haifeng;Li Bing;Lu Xueming(School of psychology,Northeast Normal University,Changchun,130024;School of philosophy and sociology,Jilin University,Changchun,130012;Cognition and Brian Science key Laboratory of Jilin Higher Education Institutes,Changchun,130024)

机构地区：[1]东北师范大学心理学院,长春130024 [2]吉林大学哲学社会学院心理学系,长春130012 [3]认知与脑科学吉林省高等学校重点实验室,长春130024

出　　处：《心理科学》2022年第3期718-724,共7页Journal of Psychological Science

基　　金：吉林省教育厅2014年规划课题(吉教科合字[2014]第B046号)的资助。

摘　　要：目前,多体素模式分析(MVPA)日渐普遍地应用于脑影像研究。近些年,机器学习的模式分类等算法在MVPA方法中被广泛应用,因其具有能够抽取高维数据模式,提高数据利用率的优点。其中一种典型的应用是利用解码的思想来解决神经表征问题,本文主要介绍了利用基于Python语言的工具库中有监督学习算法分析数据的过程。除介绍Nilearn结合Scikitlearn分析数据的步骤外,还比较不同算法的效率,为算法的选择及参数设备提供具体参考。Functional magnetic resonance imaging(fMRI)constitutes a powerful tool for addressing some basic cognitive questions in cognitive neuroscience.However,conventional fMRI analysis methods try to average across voxels that show a statistically significant response to the experimental conditions,which renders the information like the signals carried by the voxels showing a weaker response to the particular condition unavailable though it can reduce noise.Therefore,it is necessary for researchers to apply more powerful pattern classification algorithms to decode the information represented in multi-voxel activity patterns.This method is called multi-voxel pattern analysis(MVPA).At present,MVPA is increasingly used for neuroimaging data analysis,and it has become a trend to employ pattern classification and other algorithms in the field of machine learning in neuroimaging data analysis in recent years.However,the principles and applications of these tools are complex,and the neuroscience problems may not be fully considered in their development.Therefore researchers have encountered some difficulties in solving neural representation problems with these tools.In this paper,the gap between machine learning and neuroimaging was filled in by demonstrating how a general-purpose machine learning toolbox could provide stateof-the-art methods for neuroimaging data analysis while keeping the code simple and understandable by both worlds,and the analysis process of the supervised machine learning algorithm by using open data in combination with Nilearn library and scikit-learn library tools was introduced Firstly,the basic concept of machine learning and the process of data analysis,including the principle of supervised learning algorithm,was introduced in detail.Then,the use of machine learning toolbox and the selection of algorithms(including the corresponding code)matching with the software package were described to simplify the use of machine learning library and make the code simple and easy to understand through examples.W

关 键 词：机器学习 有监督学习算法 FMRI MVPA 

分 类 号：R445.2[医药卫生—影像医学与核医学] TP181[医药卫生—诊断学]