fMRI Image Analysis SPM - PowerPoint Presentation

Slide1

fMRI Image Analysis

SPM

Slide2

fMRI Processing steps

Conversion from DICOM

to

NIFTI

Organization

of

folders

Visual

Inspection

(Quality control

of a) Structural and b) Functional

images)

Pre-processing

Motion correction – Realign and

Unwarp (U)

Slice time

correction (a)

Segmentation

Get Pathname

Image Calculator

Coregister

and Estimate

Normalization and Write

(w)

Smoothing

Pre-processing flowchart

Slide3

Online tutorial Preprocessing in SPM12

https://

www.youtube.com/watch?v=Y4i1FFgWz_g

Slide4

1. Conversion from DICOM

to NIFTI

SPM12 uses NIFTI file format so you will need to convert your DICOM images into NIFTI format. You can use SPM to conver

t the images or you can use other programs such as MICROGL

https://www.nitrc.org/projects/mricrogl/

Go to folder

mricroGL and double click MRIcroGL application.Go to import convert DICOM to NIFTI.4D files can be used in SPM, FSL and AFNI. However, compressed 4D is most common in FSL (*.nii.gz), although it recognizes the non-compressed 4D file (*.nii) (FSL would then write everything in *.nii.gz). But for SPM, it has to be non-compressed, so make sure you uncheck the box “compress”.Also should check the BIDS format (Brain Imaging Data Structure) while converting.Drag the DICOM folder onto the GUI.

You can look at what the various options of dcm2nii

https://www.nitrc.org/plugins/mwiki/index.php/dcm2nii:MainPage

Slide5

2. Organization of folders

Raw_Directory

(you can keep it untouched)

Subj_001

DICOM

Task Log files

Subj_002

DICOM

Task Log files

You can use

the

subject ids with group info as prefix. For

eg.

HC – C000; MDD – D000; or you can use numbers as identifiers.

All HC starts with 1; MDD starts with 2;

rMDD

starts with 3, etc. When you come to pull subjects for groups analyses, it becomes

easier naming

all functional and structural data the same,

eg

. Resting.nii.gz, stroop.nii.gz, T1.nii.gz across all subjects, makes it easier for

batching,

t

hen

you just can loop across subjects.

NIFTI Directory

Subj_001

anat

func

dwi

Log

Subj_002

anat

func

dwi

Log

Pre-

processing

Create new folders for:

1st

level

analy

Slide6

3. Visual inspection using MRICROGL

It is very important to check your images (quality control) before performing any type of analysis.

MRICROGL is particularly useful when checking for any abnormalities in your

structural images

(T1).

For inspecting

functional images

, fslview allows you to see a video of your functional images, to search for any serious movement. Note: to view the data in mricrogl, change default by clicking on “Display” to Multiplanar, or press Ctrl + M.Select Multiplanar under Display, and then just open

your T1 by going to File open and

navigate through your data until

you reach your structural. Above is an exemple

of a good T1.

Slide7

a.

Quality Control

Structural Data

Aim is to check the Quality of the structural data

Check for any weird

abnormalities, such as big

holes, stripes,

etc.Structural MRI hardware related artifactsRF bias (B1 inhomogeneity)Non-uniform RF field causes smooth variations in intensityNeed to compensate in analysis

Slide8

a. Quality Control Structural Data

Structural MRI hardware related artifacts

RF interference

RF spikes

Ghosting

Wrap-around

Hardware/settings

Serious but uncommon, easily identifiable

Slide9

a. Quality Control Structural Data

Structural MRI

Motion related artifacts

MOTION is

v

ery

hard to

correct, so try to minimize at acquisition.

Slide10

b. Quality Control

Functional

Data

Aim is to check the Quality of the functional

data

If you can get

fslview

and play it in video mode, look for jumps, nods and stripesIf looking using mricroGL, open couple of volumes and look for stripes, artefacts, etcCheck if you have the entire brain coverageFew examples are on the next slides

Slide11

b. Quality Control Functional Data

Functional MRI artifacts

DISTORTION

Due to Bo inhomogeneity (air in sinuses)

Fieldmap

are used for correcting these

Signal Loss

Nothing we

can do, except selecting good acquisition methods, like 30deg angle, Z-shimming, etc

Phsyiological

NoiseAcquire physiological measurements Remove from your signal by covarying these measuresMOTION isvery hard to correct, so try to minimize at acquisition.

Slide12

4. Preprocessing in SPM

Preprocessing has

2 major goals: remove uninteresting variability from the data

(motion correction, slice timing, smoothing) and

to prepare the data for the statistical

analysis (Normalizing).

To prepare the data for analysis we should match all scans of an individual subject and match all subjects into a standard space. The overall goal is to increase the

quality of the images.Open SPM/batch. You find pre-defined batches inside the SPM folder, subfolder batches. To used them, open the Batch editor on the SPM for functional MRI box and load/open the SPM folder that contains a subfolder called Batches, then select the Preprocessing fMRI.m file. Then another SPM window opens. You are asked to do slice time correction and then you are asked to choose the number of sessions.

Add image of where to get the SPM batch

Slide13

4.1 Motion correction

People always move in the

scanner. Even

with padding around the head, there is still

motion.

However, it is important that every voxel corresponds to

the same anatomic point across

scans and subjects.Motion correction realigns all images to a common reference. The reference can be the first (one) image or mean of all images.Small motion (e.g. 1% of voxel size) near strong intensity boundaries may induce a 1% BOLD signal change.Each image is registered individually with the target reference image. Uses rigid body (6 DOF).

Slide14

4.1 Motion correction – Realign and Unwarp (u)

Realign

is the most basic function to match images. It is used to correct for motion during the functional scans. Uses a ridged body transformation to manipulate the scans, so it only allows translations and rotations in the x, y and z directions that are then incorporated in the

nii

files. Tries to minimize the difference between 2 scans. It can only be used in scans that have been acquired with the same pulse sequence.

Unwarp

should only be used if you believe your scans are warped. Usually realignment takes care of the distortions. When you have movements up and down and also the shape of the volume changes as a function of time, then realignment cannot take care of these changes.

Slide15

4.1 Motion correction – Realign and Unwarp (u)

Click on Data – New Session

If we have several sessions, we can preprocess them all together, by selecting different

sessions

Click on Images – navigate to the folder where your NIFTI

images are.

C

hoose all the functional files for the session or sessions. Note, on filter you can write faces.* and all the images show up. Select them all.

Slide16

4.1 Motion correction – Realign and Unwarp (u)

Change default

Num

Passes – from “Register to First” to “Register to Mean

”

Change default Interpolation – from 2nd degree interpolation to 4

th

degree

interpolation

Keep rest as default

Slide17

4.1 Motion correction – Realign and Unwarp (u)

Effect of Motion Correction

Slide18

4.1 Motion correction – Realign and Unwarp (u)

Motion Correction Output

Relative = time point to next time point – shows jumps

Absolute = time point to reference – shows jumps & drifts

Note: large jumps are more serious than slower drifts, especially in the relative motion plot

Slide19

4.1 Motion correction – Realign and Unwarp (u)

Motion Correction Remedies

Motion correction eliminates gross motion changes but assumes rigid-body motion

Other motion artefacts persist including

Spin-history changes, B0 (susceptibility) interactions & interpolation effects

Such artefacts can severely degrade functional results

Worse for stimulus correlated motion

Potential analysis remedies

Including motion parameters as

regressors

in GLM (covariates of no interest)

Removing artefacts with ICA

denoising

Outlier

timepoint

detection and exclusion (via GLM)

Excluding subjects with excessive motion

No simple cut off for “too much” motion

Slide20

PengFei

please write about the “Fixing program”

Slide21

4.2 Slice time correction (a)

Almost all

fmri

scanning takes each scan

separately, so each

slice is acquired at slightly different

times:

Data acquisition / Slice acquisitionData are acquired using pulse sequences using radiofrequency excitation followed by data collection from throughout that slice. To collect data from the entire brain, a typical pulse sequence might acquire 30 slices within a TR of 1.5 to 3.0 s depending on the scanner. Ascending/descending slice acquisition: data is acquired in a consecutive order or sequentially from one end of the imaging volume to the other.Interleaved: data is acquired in an alternating order, so data are acquired first from the odd numbered slices and then from the even numbered slices. This minimized the influence of excitation pulses on adjacent slices. A problem might be that adjacent parts of the brain are acquired at non adjacent time points within the TR. So, if slice 1 was acquired at 0 second, the 2 slice would not be acquired until 1.5 seconds later, so the same BOLD response would seem to occur earlier in the latter slice. That means that the same hemodynamic response will have different time courses within the slices. This is a problem especially in event-related designs. Temporal interpolation: Estimate the amplitude of a signal at a time point that was not originally collected, using data from nearby points. But when we do analysis, we assume that the entire brain volume is imaged at the same time, which is not true.

Slide22

4.2 Slice time correction (a)

Without any adjustment, the model timing is the same

Slide23

4.2 Slice time correction (a)

…but the timing of each slice’s data is

different

Slide24

4.2 Slice time correction (a)

Can get consistency by shifting the

data

Slide25

4.2 Slice time correction (a)

And then interpolating the data = Slice Timing Correction

Slide26

4.2 Slice time correction (a)

Slice timing correction changes the data – degraded by the interpolation

Slide27

4.2 Slice time correction (a)

Alternatively, we can get consistency by shifting the

model

Slide28

4.2 Slice time correction (a)

Batch

Click Data / New Session

Click on Dependency

Select Realign & Unwarp:

Unwarped

Images (

sess 1)As you want all the images to be slice time corrected and not only the mean image.

Slide29

4.2 Slice time correction (a)

Batch

All slice information comes from the protocol

TA – acquisition time is calculated = TR – TR/

nslices

Reference slice – usually slice that was collected at middle of the TR, here it is 45, as the slice order here was interleaved

Slice order – 1 3 5 …45 2,4,6….

44Eg to write on the slice order 1:3:45 2:4:44

Slide30

4.3 Segmentation

Is used to separate the distinct tissues such as grey

matter (GM),

white matter

(WM) and

CSF in anatomical scans. This can also be used for normalization. This step is performed with the anatomical image (T1-weighted).

Slide31

4.3 Segmentation

Batch

Click Volumes

Navigate to T1 NIFTI folder and select your T1 image

Change default Save bias corrected – from “save nothing” to “save field and corrected”

Change default native tissue – from “native space” to “native +

Dartel

imported”Change default Warped tissue – from “none’ to “modulated + unmodulated”***last two changes are because these files are useful if we plan to do structural analyses later. **change the last defaults for all tissue probability map (1 – 6)

Slide32

4.3 Segmentation

Batch

Change default Deformation Field - from “None” to “inverse + forward”

This step moves segmented grey matter into the MNI space (ICBM space template – European brains)

Forward deformation – represents the transformation needed to move subject space to MNI space

Inverse deformation – represents the transformation needed to move MNI space to subject space

Slide33

4.4 Get Pathname

This step is really only to get the pathname of your T1 folder where segmentation outputs are

saved.

Slide34

4.5 Image calculator

This step put together the WM+GM+CSF images into one and corrects for bias field by multiplying this added image by bias corrected image (output from segmentation step

)

It is just to create a mask corrected for intensity changes, so co-registration with functional images are more accurate. It is not needed for normalization, as deformation filed is already created in segmentation step using T1 (GM)

Slide35

4.5 Image calculator

Input images – c1 (GM), c2 (WM), c3 (CSF) and bias corrected

All of them are outputs of segmentation steps, so we can

choose using dependency

Output name – T1_bias_brain

Output

Directory: Directories Unique

Expression = (i1 + i2 + i3).* i4 Adding WM, GM, CSF And multiply with bias corrected image Leave all the rest as default.

Slide36

4.6 Coregister and Estimate

Coregistration

is done to match scans of different modalities (

eg

., T1 and T2) and like realignment it only allows rigid body

transformations.

Estimation

determines the transformation parameters without changing the bitmap. It incorporates them into the nii file. As a reference image you should select the scan to which another scan is going to be compared. As a source image, you should select the scan that is going to be transformed onto the reference scan. In this step we take care of the connection between structural and functional images within the subject. 

Slide37

4.6 Coregister and Estimate

Coregister

mean functional image

(

Unwarped

mean, source image)

to T1_bias_brain (

reference image) Other images, slice timming corr Images session1. Use dependency to choose these images.It moves functional images in the space registration as T1. 

Slide38

4.7 Normalization and write (W)

Functional

During

the normalizing step we connect our functional

data

with the normative data from MNI.

This

step requires smoothing. Importantly, you should consider using a “bounding box” which defines the amount of MNI space that will be incorporated in your new files.StructuralDuring the normalizing step we connect our structural data with the normative data from MNI. This step does not require smoothing.  

Slide39

4.7 Normalization and write (W)

Move functional data and T1_bia_brain into MNI

space (use dependency in these steps).

Need to do in two Normalize – Write

steps (functional on the left, structural on the right

).

 

Slide40

4.8 Smoothing

Smoothing is one of the last steps and aims at blurring the functional images to correct for any remaining functional and anatomical differences between subjects.

Nevertheless

the more you smooth the less resolution you get.

If

you are interested in small regions you should not smooth that much.

 

Smooth data –rule of thumb twice as your voxel size

Slide41

Pre-processing flowchart

REALIGN & UNWARP

SLICE TIMING CORRECTION

Preparing Structural Data

SEGMENTATION

GM (C1)

WM (C2)

CSF (C3)

Bias Corrected

Forward + Inverse Deformations

Create Bias corrected T1 mask

Realigned Images

Realigned Mean Image

Combining

func

+

struc

Preparing functional data

CO-REGISTRATION

Source

Reference

Other

Co-registered STC images (not saved)

NORMALIZATION (MNI)

Forward Deformation

Normalized functional Images

SMOOTHING

NORMALIZATION (MNI)

Forward Deformation

Normalized T1_bias_brain Images

1

2

3

4

5

6

7

8