2. Setting Up
cd ~/Desktop/Exampledata/
fslinfo 052212_s09_dti_eddy_fm.nii
eddy_correct 052212_s09_dti_eddy.nii 0 (correcting eddy
current; registering image w.r.t. image 0 or the initial
image while processing each and every image)
bet 052212_s09_dti_eddy_fm.nii
052212_s09_dti_eddy_eddy_fm_strip.nii –f 0.3
4. BET (Brain
Extraction
Tool)
Removes skull and meninges from
anatomical scan
Input anatomical dataset
_brain added in input data name
defaults ok
GUI activities shown in command
line
observe in FSLView
Only input image location
insertion is important
6. Misc
Most as default
Brain /background
threshold, % 10 (Whether
part of brain/ background
based on fraction of
maximum intensity voxel.
If 100% , anything less
than 10% is not brain.)
Noise level can be
estimated from 4D data.
(Default)
7. Data
Allows to determine different
parameter of 4D datasets, can
have multiple datasets to be
analyzed. Each analyzed same
way (time, sequence and
condition).
Input = 1
Select 4D data (to load data)
High pass cut-off = 100s
(cycles greater than 100 sec
to be removed)
Default: Total volume (loads
itself); delete volumes
(usually 0); TR(s) (sometimes
has to be changed).
8. Pre-stats
MCFLIRT (FSL traditional motion
correction algorithm) (tries to match
all volume to middle volume for that
run)
B0 unwrapping (only if interested to
study areas of brain near to signal
drop-out. Some examples: cortical
patches near sinuses or pre-frontal
cortex) If field-map is acquired,
phase on top and magnitude image on
bottom. Field-map is not usually
recommended.
Slice timing correction is important
(as all slices are not acquired at
once, regular up(bottom slice first
in z direction,…): Interleaved (even
and odd)
BET (brain extraction for functional
images) (on is ok for future co-
registration)
Spatial smoothing (averaging data
from nearby voxels, important for
SNR)
Intensity normalization (match
average intensity to certain preset)
(Off)
Temporal filtering option (high pass
filter only for perfusion MRI)
Melodic ICA (decompose data to
several independent component) (to
remove noise and isolate certain
components)
9. Stats
Set up model that can be
used in each voxel in
dataset
How time file should be
covered to be read into FEAT
(default is ok)
FILM pre-whitening (default
: ON) (try to auto-correlate
in same image)
Don’t Add Motion Parameters
(only if patient moved a
lot) and text file with time
information
10. Full Model
Setup (General
Linear Model)
EV (explanatory variable)
◦ Number of original EVs : 4 (Left, Right, Breathe,
Hold breathe) (4 condn/run)
◦ EV name: TapLeft
◦ Basic Shape : Custom (3 column format)
◦ Filename: FSL_onset/FSL_TapLeft
◦ Convolution: Gamma
◦ Do not orthogonalize, Add temporal filtering,
apply temporal filtering (default ok )
Contrasts & F-tests
◦ Contrasts : 8
◦ Fill in like the attached image
View design (After you click, you will
observe a figure with 8 columns, Left, Left
with temporal derivative and so on) (First
red line represent 100 sec, if any activity
lasted more than that, the high pass filter
will cut-off.)
Observe
11. What typical time file
looks like?
-----Command line-----
ls
vi FSL_TapLeft_1.txt
Second column
15
15
15
15
For how long the
condition was
present
First
column
46.569
125.569
201.569
278.569
Time of
occurrence of
activity in
sec
Third column
1
1
1
1
(Parametric
modulation) Height
of BOLD signal
12. View
design
After you click, you will
observe a figure with 8
columns, Left, Left with
temporal derivative and so
on
First red line represent
100 sec, if any activity
lasted more than that, the
high pass filter will cut-
off.
13. Efficiency
Design efficiency image
First box: Correlation
between different
regression factors.
Diagonals refers to auto-
correlation, so are white.
Second square: Efficiency
in estimating each of
these regresses. If any of
diagonal very dark, may
have issues.
Image bottom shows
percentage effect
required.
14. Post-stats
Usually correct for multiple
comparisons.
Pre-threshold masking (if
any specific hypothesis)
(boosting statistical power)
Thresholding (Uncorrected,
Voxelwise selection,
clusters after z-maps
selected) can appear at this
p-threshold (0.05)
Contrast masking (mask based
on another contrast) (create
time series plot and compare
to actual model fit)
15. Registrati
on
Potential spatial correlation
during analysis. (Usually done at
last)
Co-registration (all to same
anatomical space)
Main structural image
◦ anatomical image with all non-
brain matter s007a01001.nii.gz
Standard space (7 different
templates provided)
◦ /usr/local/fsl/data/standard/MNI
152_T1_2mm_brain (Montreal
Neurological Institute resampled
to 2mm isotropic voxels)
Linear: Full search /12 DOF (Do not
non-linear)
Save as analysis.fsf files
Go
16. Output directory: output
Progress pops-up in html
Different tabs (Prestats, Stats, Post-stats (rendered brain
slices, minimum 2.3 or higher), Registration, Log(track of
everything))
Wait for analysis to be finished to change tabs, error or any
sort of motion registered is observed here.
Time series plot
Co-registration (red lines show grey-white matter boundaries)
17. Actual FEAT directories
---- Command Line -----
ls
output.feat s005a001.nii.gz s006a001.nii.gz
cd output.feat/
ls (shows all files) (Folders: logs, mc(motion correction),
reg(registration))
cd stats
ls
fslview (open : example_func, add:
output.feat/stats/zstats/zstat1.nii.gz)
Threshold in fslview on top change to (Min 3 Max 12)
18. Combining and automating
FEAT
Second run with all same parameters
FEAT
DATA TAB: o/p directory : output_subj_01
STATS TAB : Load new text files (Reload FEAT)
Higher Level Analysis (Inputs are lower level FEAT directories)
New window (Top: output.feat / Bottom: output_run2.feat) OK
GO
New results are in individual folder for each contrast
FSLview: open /usr/local/fsl/data/standard/avg152T1_brain add
cope1/stats/zstat1.nii.gz (Thres: 1-3)
19. Automating FEAT
Each run, output.feat is generated.
load output.feat > design.fsf
mkdir tmp
cp output.feat/design.fsf tmp
cd tmp
ls
vi design.fsf (open briefly) (everything in short .txt format)
changes at certain brief points
wq
gedit design.fsf
20. Observing
the FA
maps in
FSL
Open FA map in FSLview
and then add V1 map
From the information
dialog ((i) at bottom),
change image type to
Diffusion Tensor, display
as to RGB and modulation
to dti_pt4_FA.