Spatial Normalization

There are two commonly used neuroimaging reference spaces: Talairach and MNI. Below are documents on the creation of the Talairach space, and the differences between the two. Meta-analysis and other data mining is dependent upon the coordinates of reported papers being correct and all within the same brain reference space. Figuring out which reference space is used when papers report coordinates is both very important and tricky. Pay close attention to "template" and "reference space" in papers you are coding.

The Brain Template video provides an overview of reference spaces.

The Brett and icbm2tal transforms are available within GingerALE's "Transform Coordinates" menu item.

Text from grant

Specifically, BrainMap data are limited to those that report statistical parametric maps (SPMs) computed in a voxel-wise manner over the whole brain in a standardized anatomical coordinate space, as first described by Fox, Friston and others (Fox et al 1985; Fox et al 1988; Friston et al. 1991).

Scribe Analysis – Training Mode Text

Software Package - The analysis software that was used to spatially normalize the subjects’ raw data. This is also known as “alignment”, “registration”, or “co-registration”. If not mentioned, use Unknown.

Template Brain - Subjects’ images are often spatially normalized by fitting them to a widely known high-resolution brain image. The most widely known is probably ICBM152, which is the default template for all versions of FSL and most versions of SPM. (SPM94 through SPM99 had many changes to the default. SPM2 through SPM12 use ICBM152 as the default.) If you can infer which template was used, do so. Otherwise, use Unknown.

Alternative to using a template, images can be registered to Talairach space using landmark registration. In this case, note which version of the Talairach atlas is being used. (Tal 1967 vs Tal 1988). If an atlas version is not mentioned, use Unknown.

Transform - It’s not uncommon to perform the analysis in MNI space, perhaps using FSL or SPM, and convert the results to Talairach space using a transformation matrix. The most common transforms are “mni2tal”, also known as “Brett”, and “icbm2tal”, also known as “Lancaster”.

Many papers use no transform at all and report coordinates in the same space as the analysis software.

If it looks like the data as in MNI space at one point and Talairach space at another, look carefully for any reference to transforms, registration or re-alignment. If you think a transform was applied, but cannot be sure which one it was, use “Unknown”.

Reference Space - Nearly every paper should state if published coordinates are in Talairach space or MNI space.

If not, you may be able to use other analysis information to determine which reference space the coordinates are in. Two very popular software packages, FSL and SPM, default to MNI space. If the templates MNI305 or ICBM152 are used, it’s almost certainly in MNI space. If a transform was used the most likely analysis was done in MNI space and reported coordinates are in Talairach.

If the reference space is still unknown, or if there is conflicting information about the reference space, make a note of it in the Feedback panel.

From Analysis:

Registration Software - The analysis software used to align the images to a well-known reference space.

Brain Template - The brain template used to align images to a common stereotactic space. If aligned to Talairach space without a template, code which Talairach atlas was used.

Transform - If the aligned data was transformed to another coordinate space (usually MNI -> Talairach), record the transform used here.

If MNI coordinates are reported in Talairach and the transform is not specified in the text, indicate the transform is "Unknown" and make a note in the short description.

All MNI coordinates will be automatically converted to Talairach space using the icbm2tal transform (Lancaster et al., 2007). All MNI coordinates converted to Talairach space via the Brett transform in the original publication will be subject to 2 transforms: (1) reverse-Brett to convert back to MNI space and (2) icbm2tal for correct transformation from MNI to Talairach space.

Reference Space - The standard coordinate space used when reporting coordinates. The analysis software, brain template and transforms can be used to infer which reference space was used. If reference space is unknown or if fields are contradictory, make a note of it in the short description.

From Template Video Script

Into

show surface in Mango, rotate + resize a bit

Brain Templates, okay. So we’re talking about how the individual subjects’ brain scans are aligned to make them more directly comparable. Differences can be due to the resolution of the scanner, variations in positioning within the scanner or differences in the subject’s brain size. Most of these can be accommodated for with some rotation, resizing, shifting. This re-alignment is called registration, or co-registration, or transforming or applying a transform.

Transforms

show icbm_spm2tal.m

The “transform” part is in reference to a transformation matrix, which describes exactly how you’re shifting, resizing and rotating. One transform looks like this. If I remember my Matrix Algebra, the diagonals are resizing (So this is shrinking about 7% in the x and y directions and a little more than that in Z), around them are rotating (I think this is rotating in the direction of lifting your nose up?) and the right-most column is shifting (1mm to the left, almost 2 to the back and 4mm upwards).

References

But how do you know which transform to apply? You can choose one subject and use it as a “reference brain” or a “template” and find the transform that best aligns each subject to it. You could choose the subject that would require the smallest transforms for the other subjects to match to it. Or you could create a template by finding the average brain. Either of those would work fine within a single study, but it would be difficult to compare your results to other studies since there’s no guarantee that your reference brain would line up with others’.

Two Standards

So to help compare neuroimaging results across studies, we have two commonly used reference spaces:

MNI space (from the Montreal Neurological Institute) and Talairach space (from the Talairach and Tournoux atlas)

MNI

MNI space is a little easier to explain, because it uses a template as we were just describing. Popular image processing software, such as FSL and SPM, incorporate a template from the Montreol Neurological Institute, (such as MNI 152 or MNI 305). Each subjects’ image data can be transformed using an automated over-all best-fit alignment with the included template.

The template includes a coordinate space, so regions can be discussed using (x,y,z) coordinates.

Talairach

(*Originally - not so true anymore) Talairach space, on the other hand, is a reference space which focuses on the alignment process instead of a template brain. Registering a brain to Talairach space requires finding several anatomical landmarks. The first is the mid-sagittal plane, which runs between the two hemispheres. The Y-Z plane is oriented along the mid-sagittal plane. Next, the Y axis is aligned along the “AC-PC line” which includes several midbrain landmarks (bottom of the AC, thalamus, … top of cerebellum). The origin is set to the AC. After all the rotation and translation is finished, scaling is applied to match the extremities of the brain to a fixed bounding box size. This template is a brain that was first aligned in MNI space, then registered to Talairach space.

What’s the difference?

MNI brain is larger: show edge differences (mostly at top, but scroll through with Link Nav on)

MNI origin is different: jump to origin on both with ‘O’ in Mango, show they are a few mm off

So MNI vs Tal will have differences both in midbrain and cortical regions

Talairach space was invented first, in the 80s, and could be used without a template brain, but required anatomical expertise. and the included template brain was an individuals, assumed to not be representative

MNI space arrived in the 90s with SPM96, and was very easy to use with automated registration, but was not really comparable with Talairach space (used different landmarks for registration, AC is 4mm below origin/nose tilted down, larger bounding box). the included template was an average, assumed to be more representative, but also much larger than the average brain.

(Side note: make sure to always use “World Space” in Mango if you want transforms applied. Otherwise it will be “Image Space” which puts the origin in one of the corners and the sizes depend on the image resolution.)

Why is it important?

meta-analysis and other forms of data-mining group together data from both reference spaces and relies on agreement between foci to find results. a shift of a few millimeters in half our data could have a pretty large negative effect.

What can we do?

We have some transforms available that can convert between these reference spaces pretty well. This one from before is for converting from SPM’s MNI to Talairach. Why do I say SPM’s MNI? Well, it was found that FSL and SPM register brains to MNI space slightly differently. There’s a transform for each software package, and a combined general MNI to Talairach transform. And you can use the inverse transform of any of these for Talairach to MNI. These transforms were published by Lancaster et al in <> and are often called “icbm2tal” or “Lancaster” transforms.

Historical Note

Before icbm2tal transforms were developed, the transform usually used was the Brett transform shown here. In Lancaster’s paper, it was shown that the Brett transform introduced more error (*gave different results) than the new transforms, so when a paper using the Brett transform is submitted to BrainMap, we undo the Brett transform to convert it back to MNI and then apply an icbm2tal transform.

Coding

So when you’re coding a paper, this is why we ask for the brain template used for the published coordinates (usually Tal or MNI), software package used in the image registration (usually FSL or SPM) and any transforms applied (usually Brett or icbm2tal or none).

Training Video

Copyright © 2003 Research Imaging Institute. All rights reserved.