FSLNet Workflow

FSLNet 是基于Matlab开发的网络分析工具箱，默认采用的input是fsl melodic 的组ICA输出结果，基于官方教程总结下面的workflow。

rfMRI FSLnets Workflow

Input Preparation

Define nodes

(spatial maps)

melodic -i input_files.txt -o groupICA100 \ --tr=0.72 --nobet -a concat \ --bgimage=$FSLDIR/data/standard/MNI152_T1_2mm_brain.nii.gz \ -m $FSLDIR/data/standard/MNI152_T1_2mm_brain_mask.nii.gz \ --report --Oall -d 30

Extract timeseries from nodes

dual_regression groupICA100/melodic_IC 1 -1 0 groupICA100.dr cat input_files.txt

create images of nodes

slices_summary groupICA100/melodic_IC 4 $FSLDIR/data/standard/MNI152_T1_2mm groupICA100.sum -1

Matlab setup

Package Requirements

Offical toolbox

Statistics

is a must

Bioinformatics

MATLAB's SVM

Signal Processing

see timeseries

LIBSVM

use the LIBSVM implementation of SVM for netmat-based classification

L1precision

to estimate L1-norm regularised partial correlation matrices

Pairwise causal

look at causal directionalities

FSLNets

下载FSLNets包 http://www.fmrib.ox.ac.uk/~steve/ftp/fslnets.tar.gz

network_analysis.m

Official template is in FSLNets folder, name is 'nets_examples.m'

Script setup

add package paths

load group map/ dual regression output directory

check mean timeseries spectra

Node Cleaning

set up node list

ts.DD

good node

ts.UNK

optional
for unknown components

regress the bad node out

check cleaned data and visualization

Calculating netmats for each subject

netmats0= nets_netmats(ts,0,'cov'); % covariance (with variances on diagonal) netmats0a= nets_netmats(ts,0,'amp'); % amplitudes only - no correlations (just the diagonal) netmats1= nets_netmats(ts,1,'corr'); % full correlation (normalised covariances) netmats2= nets_netmats(ts,1,'icov'); % partial correlation netmats3= nets_netmats(ts,1,'icov',10); % L1-regularised partial, with lambda=10 netmats5= nets_netmats(ts,1,'ridgep'); % Ridge Regression partial, with rho=0.1 netmats11= nets_netmats(ts,0,'pwling'); % Hyvarinen's pairwise causality measure

Group-average netmat summaries

group mean

output:[t-test Z value, Mean value]

group average network hierachy

Visualization

interactive web-based display

Cross-subject comparison

Calculating significant group difference

two-ample t-test

output

corrected p
uncorrected p

Displaying significant group differences

displaying boxplot

Multivariate cross-suvject analysis

ML-based classification

overall pattern change between two groups