[综述笔记]Flexible large-scale fMRI analysis: A survey

论文网址：Flexible large-scale fMRI analysis: A survey | IEEE Conference Publication | IEEE Xplore

英文是纯手打的！论文原文的summarizing and paraphrasing。可能会出现难以避免的拼写错误和语法错误，若有发现欢迎评论指正！文章偏向于笔记，谨慎食用

1. 省流版

1.1. 心得

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

2.2. Introduction

2.3. FMRI Data Analysis Methods

2.3.1. Independent Component Analysis

2.3.2. Dictionary Learning

2.4. Group Analysis Methods

2.4.1. ICA for Group Analysis

2.4.2. DL for Group Analysis

2.5. Large-Scale FMRI Analysis: Challenges and Opportunities

2.5.1. Independence or Sparsity?

2.5.2. Multi-Site FMRI Data Sharing

2.5.3. Decentralized Analysis

2.6. Conclusion

3. 知识补充

3.1. blind source separation (BSS)

4. Reference List

1. 省流版

1.1. 心得

（1）一不小心考古了，算了，b类会议看就看吧

（2）开幕雷击，这abstract怎么第二句就上方法啊。“功能性磁共振成像（fMRI）为大脑提供了一个窗口，在研究甚至临床环境中被广泛采用。数据驱动的方法，例如基于潜在变量模型和矩阵/张量分解的方法，正越来越多地用于 fMRI 数据分析。”来得太陡了吧xd。我都不知道这是啥

（3）说fMRI是大脑的一个非入侵窗口哈哈哈哈哈哈写论文还是不要乱比喻吧

（4）接ROI先验！！

（5）为什么又是机器学习...

1.2. 论文总结图

2. 论文逐段精读

2.1. Abstract

①Data-driven methods such as latent variable models and matrix/tensor factorizations based methods can be used in fMRI analysis.

②Challenge: the processing of fMRI data

2.2. Introduction

①Functional connectivity in rest state exposes the spontaneous cognitive process of the brain

②Hypothesis-driven analyzing methods: region-of-interest (ROI) or seed-based（啥玩意儿，长知识了）.⭐However, they are hypothesis-driven, which means they are priori

③Data-driven analyzing methods: this method is for blind source separation (BSS). Independent component analysis (ICA) can divide brain regions with consistent time without overlapping without prior experience. Moreover, dictionary learning (DL), a matrix decomposition method, can also caputure the sparsity characters of fMRI data

vascular adj.血管的;维管的;脉管的

hemoglobin n.血红蛋白

oscillation n.振动;摆动;振幅;摇摆;(情感、行为、思想的)摇摆不定，变化无常，犹豫不定;浮动;一次波动

2.3. FMRI Data Analysis Methods

①The temporal matrix $\mathbf{A}$ and spatial matrix $\mathbf{S}$ of fMRI data:

where the volumn is voxel

2.3.1. Independent Component Analysis

①我实在是没看懂这一堆定义表达式

2.3.2. Dictionary Learning

①Function of DL:

$\mathbf{x}(v)=\mathbf{A}\mathbf{s}(v)+\mathbf{e}(v),v=1,2,\ldots,V$

where $\mathbf{s}(v)$ is sparse hypothetically and $\mathbf{e}(v)$ denotes the noise

parsimonious adj.吝啬的;小气的;悭吝的

postulate vt.假设;假定 n.假定;假设

tantamount adj.等于;等同的;无异于;效果与…一样坏

2.4. Group Analysis Methods

2.4.1. ICA for Group Analysis

①Function of IVA:

$\mathbf{X}_k=\mathbf{A}_k\mathbf{S}_k^\mathrm{T}(\mathrm{or},\mathrm{~x}_k(v)=\mathbf{A}_k\mathbf{s}_k(v),\mathrm{~}v=1,\mathrm{~}\ldots,V)$

where the $\left \{ \mathbf{A}_k \right \}$ denotes individual time courses and $\left \{ \mathbf{S}_k \right \}$ denotes spatial maps;

$\mathbf{X}_k\in\mathbb{R}^{N_k\times V}$ denotes fMRI data, $k=1,...,K$ is subject

②Integration of IVA to SCV:

③Receiver operating characteristic curves for IVA and group ICA:

2.4.2. DL for Group Analysis

①Function of DL:

$\mathbf{X}_k=\mathbf{A}_k\mathbf{S}_k^\mathrm{T}+\mathbf{E}_k,\mathbf{S}_k=\mathbf{\bar{S}}+\mathbf{F}_k,k=1,2,\ldots,K$

where $\mathbf{E}_k$ and $\mathbf{F}_k$ represent Gaussian perturbations

2.5. Large-Scale FMRI Analysis: Challenges and Opportunities

2.5.1. Independence or Sparsity?

①Some reseachers reckon that ICA mostly focuses on sparsity

②IVA is able to capture the mixing matrices and spatial maps, but costs high comuputing time

2.5.2. Multi-Site FMRI Data Sharing

①The unification and sharing of neuroimaging data are subject to many limitations and obstacles

②⭐The authors believe that the datasets of various institutions do not need to be uniformly stored on a single website, but rather stored separately. However, they should have a unified local analysis model. For instance, ViPAR, ENIGMA, and COINSTAC have already implemented these.

consortium n.联盟;(合作进行某项工程的)财团，银团，联营企业（复数是consortia）

post-hoc 因果的

2.5.3. Decentralized Analysis

①Fidelity, the amount of information exchange between repositories, the parallelism of scalability processing, and privacy/policy compliance of data sharing are all issues that decentralization methods need to consider

fidelity n.忠诚;忠实;(对丈夫、妻子或性伴侣的)忠贞;准确性;精确性

2.6. Conclusion

This article focuses on the analyzing methods of fMRI data, mainly containing independent and sparse matrix factorization approaches.

3. 知识补充

3.1. blind source separation (BSS)

（1）定义：

盲源分离（Blind Source Separation，BSS）是一种技术，旨在从一组混合信号中分离出各个未知源信号。在未知系统的传递函数、源信号的混合系数及其概率分布的情况下，该技术仅利用源信号之间相互独立这一微弱已知条件来实现分离。盲源分离技术在信号处理、通信、音频处理等领域有广泛应用。

在数学描述上，盲源分离通常涉及假设N个统计独立的未知信号经过未知信道的传输后，由M个传感器检测获得M个观测信号。盲源分离问题的目标是找到一个分离矩阵，使得观测信号通过该矩阵后能够尽量完全地分离出源信号的各个组成。

盲源分离的实现方法多种多样，其中一些方法利用了信号的统计特性或信号之间的独立性。在实际应用中，传感器测得的信号通常是源信号及其延时信号的混迭，这被称为卷积混迭。针对这种情况，盲反卷积方法可以通过观测信号估计信道冲激响应，进而恢复源信号。

（2）主要方法：ICA、PCA

4. Reference List

Kim, S., Calhoun,V. & Adalı, T. (2017) 'Flexible large-scale fMRI analysis: A survey', 2017 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). doi: 10.1109/ICASSP.2017.7953372

本文来自互联网用户投稿，该文观点仅代表作者本人，不代表本站立场。本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如若转载，请注明出处：http://www.mzph.cn/news/770216.shtml

如若内容造成侵权/违法违规/事实不符，请联系多彩编程网进行投诉反馈email:809451989@qq.com，一经查实，立即删除！