Co-registration and Spatial Normalisation Martin Chadwick and Catherine Sebastian.

Co-registration and Spatial Co-registration and Spatial NormalisationNormalisation

Martin Chadwick and Catherine Martin Chadwick and Catherine SebastianSebastian

OverviewOverviewOverviewOverview

Motioncorrection

Smoothing

kernel

(Co-registration and) Spatialnormalisation

Standardtemplate

fMRI time-series Statistical Parametric Map

General Linear Model

Design matrix

Parameter Estimates

Preprocessing StepsPreprocessing Steps• RealignmentRealignment

– Motion correction: Adjust for movement Motion correction: Adjust for movement between slicesbetween slices

• CoregistrationCoregistration– Overlay structural and functional images: Link Overlay structural and functional images: Link

functional scans to anatomical scanfunctional scans to anatomical scan• NormalisationNormalisation

– Warp images to fit to a standard template Warp images to fit to a standard template brainbrain

• SmoothingSmoothing– To increase signal-to-noise ratioTo increase signal-to-noise ratio

• Extras (optional)Extras (optional)– Slice timing correction; unwarpingSlice timing correction; unwarping

• Refers to any method for realigning Refers to any method for realigning imagesimages– Realignment for motion correction (last Realignment for motion correction (last

week)week)– Aligning or overlaying images from Aligning or overlaying images from

different modalitiesdifferent modalities

Co-registrationCo-registration

T2* EPI image (low

resolution)

T1 structural MR image

(high resolution)

Why Co-register structural and Why Co-register structural and functional images?functional images?

• Can overlay functional activations Can overlay functional activations onto an individual’s own anatomyonto an individual’s own anatomy

• Can overlay group-level functional Can overlay group-level functional activations onto an average structuralactivations onto an average structural

• Gives you a better spatial image for Gives you a better spatial image for later normalisation step, as warps later normalisation step, as warps derived from the higher resolution derived from the higher resolution structural image can be applied to the structural image can be applied to the functional imagefunctional image

Recap: realignment Recap: realignment parametersparameters

• Like motion correction (realignment), Like motion correction (realignment), co-registration makes use of 6 co-registration makes use of 6 parameters…parameters…

YawPitch

Roll

ZX

Y

Translation Rotation

Differences between Differences between realignment and co-realignment and co-

registrationregistration• Images may not be quite the same Images may not be quite the same

shape (distortion of EPI images, shape (distortion of EPI images, especially in phase encode direction)especially in phase encode direction)

• Structural and functional images do Structural and functional images do not have the same signal intensity in not have the same signal intensity in the same areas. Therefore there are the same areas. Therefore there are additional steps additional steps No direct voxel to

voxel match

The Normalised Mutual The Normalised Mutual Information ApproachInformation Approach

• Different material will have different Different material will have different intensities intensities withinwithin a scan modality (e.g. a scan modality (e.g. air will have a consistent brightness, air will have a consistent brightness, and this will differ from other and this will differ from other materials such as white matter).materials such as white matter).

• When looking between modalities, When looking between modalities, these consistencies can be used to these consistencies can be used to compare images compare images

An exampleAn example

aligned not aligned

Joint histogram shows little

noise

More noise: hard to define

structures with certainty

Additional pointsAdditional points

• In some studies structurals are not In some studies structurals are not taken – it is possible to conduct fMRI taken – it is possible to conduct fMRI analysis without co-registering to a analysis without co-registering to a structuralstructural

• Sometimes when you co-register, Sometimes when you co-register, you have to reslice the datayou have to reslice the data- E.g. change image dimensions from 3x3x3 to

2x2x2, or change apparent direction of data collection from axial to coronal

- Useful if two images have very different voxel sizes

- Often involves interpolation

- Often used with PET data

Co-registration in SPMCo-registration in SPM

Co-Co-registration registration in SPMin SPM

Make selection

Explains each option

Template: image that remains stationaryImage that is ‘jiggled about’ to match templateDefaults used by SPM for estimating the match, including Normalised Mutual InformationReslice options: choose from the menu for each of the three options (usually just defaults)

Run

Preprocessing StepsPreprocessing Steps• RealignmentRealignment

– Motion correction: Adjust for movement Motion correction: Adjust for movement between slicesbetween slices

• CoregistrationCoregistration– Overlay structural and functional images: Link Overlay structural and functional images: Link

functional scans to anatomical scanfunctional scans to anatomical scan• NormalisationNormalisation

– Warp images to fit to a standard template Warp images to fit to a standard template brainbrain

• SmoothingSmoothing– To increase signal-to-noise ratioTo increase signal-to-noise ratio

• Extras (optional)Extras (optional)– Slice timing correction; unwarpingSlice timing correction; unwarping

Matthew Brett

What is Normalisation?

Warps images from Warps images from different participants different participants onto a template brainonto a template brain

• Improve the statistical power of the analysisImprove the statistical power of the analysis

• Generalise findings to the population levelGeneralise findings to the population level

• Identify commonalities and differences Identify commonalities and differences between groups (e.g. patient vs. healthy)between groups (e.g. patient vs. healthy)

• Report results in standard co-ordinate system (e.g. Talairach)

Why Normalise? Why Normalise?

We can average the signal across participants, allowing us We can average the signal across participants, allowing us to derive group statistics. This can allow us to:to derive group statistics. This can allow us to:

SPM: Spatial NormalisationSPM: Spatial Normalisation• SPM uses a voxel-intensity-based approach to SPM uses a voxel-intensity-based approach to

normalisation.normalisation.

• adopts a two-stage procedure :adopts a two-stage procedure :

• Step 1:Step 1: Linear transformation (12-parameter affine). This Linear transformation (12-parameter affine). This step accounts for the major differences in head shape and step accounts for the major differences in head shape and position, but there will be remaining smaller-scale position, but there will be remaining smaller-scale differences.differences.

• Step 2:Step 2: Non-linear transformation (warping). The non-linear Non-linear transformation (warping). The non-linear step is designed to take care of the smaller-scale step is designed to take care of the smaller-scale differences in brain anatomy.differences in brain anatomy.

Alternatives – anatomy based approaches e.g. FreeSurfer

Step 1: Affine Step 1: Affine TransformationTransformation• Determines the Determines the

optimum 12-optimum 12-parameter affine parameter affine transformation to transformation to match the match the sizesize and and positionposition of the images of the images

• 12 parameters = 3 12 parameters = 3 translations and 3 translations and 3 rotations (rigid-body) rotations (rigid-body) + 3 shears and 3 + 3 shears and 3 zoomszooms

Rotation Shear

Translation Zoom

Step 2: Non-linear Step 2: Non-linear RegistrationRegistration

• The model for defining nonlinear warps uses deformations consisting The model for defining nonlinear warps uses deformations consisting of a linear combination of low-frequency periodic basis functions.of a linear combination of low-frequency periodic basis functions.

Affine registration

Templateimage

Non-linearregistration

withoutregularisation.

Non-linearregistration

usingregularisation.

Over-fitting and RegularisationOver-fitting and Regularisation

CaveatsCaveats• Impossible to make a meaningful perfect structural match between Impossible to make a meaningful perfect structural match between

subjects, due to individual differences in anatomysubjects, due to individual differences in anatomy

• Even if anatomy is well-matched, it does not guarantee that functionally Even if anatomy is well-matched, it does not guarantee that functionally homologous areas are spatially aligned – we don’t know the extent to homologous areas are spatially aligned – we don’t know the extent to which individuals may vary in their structure-function relationships.which individuals may vary in their structure-function relationships.

• Pathology creates a particular problem here, as even relatively confined Pathology creates a particular problem here, as even relatively confined abnormalities or lesions can cause mis-registration in widespread areas of abnormalities or lesions can cause mis-registration in widespread areas of the brain, due to the global nature of the normalisation process. Need to the brain, due to the global nature of the normalisation process. Need to bare this in mind in patient studies.bare this in mind in patient studies.

Solution Solution

• A partial solution for any remaining small-scale differences in anatomical or A partial solution for any remaining small-scale differences in anatomical or functional location is offered by the next stage of pre-processing, where functional location is offered by the next stage of pre-processing, where the images are spatially smoothed. the images are spatially smoothed.

Normalisation in SPMNormalisation in SPM

Select OptionSelect Option

Select image to be matched to templateSelect image to be matched to template

Select image(s) to be warped using the Select image(s) to be warped using the sn.mat calculated from the Source sn.mat calculated from the Source ImageImage

Select SPM template:Select SPM template:Structural – spm5\templates\T1.niiStructural – spm5\templates\T1.niiFunctional - spm5\templates\EPI.niiFunctional - spm5\templates\EPI.nii

Select voxel sizes for warped output Select voxel sizes for warped output imagesimages

Sources:Sources:• Friston, K. J. Introduction: Experimental design and statistical Friston, K. J. Introduction: Experimental design and statistical

parametric mappingparametric mapping

http://www.fil.ion.ucl.ac.uk/spm/doc/intro/http://www.fil.ion.ucl.ac.uk/spm/doc/intro/

• Ashburner & Friston “Rigid Body Registration” Chapter 2, Ashburner & Friston “Rigid Body Registration” Chapter 2, Human Brain Function, 2Human Brain Function, 2ndnd ed.; ed.; http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/

• Ashburner & Friston “Spatial Normalization Using Basis Ashburner & Friston “Spatial Normalization Using Basis Functions” Chapter 3, Human Brain Function, 2Functions” Chapter 3, Human Brain Function, 2ndnd ed.; ed.; http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/http://www.fil.ion.ucl.ac.uk/spm/doc/books/hbf2/

• Rik Henson’s Preprocessing Slides:Rik Henson’s Preprocessing Slides: http://imaging.mrc-cbu.cam.ac.uk/imaging/ProcessingStreamhttp://imaging.mrc-cbu.cam.ac.uk/imaging/ProcessingStream

• Previous MfD SlidesPrevious MfD Slides