Step 10: Registration and Spatial Alignment

Overview

Registration aligns images from different spaces so that the same brain structures overlap. In DTI preprocessing, you need to create a chain of transformations that connect three spaces:

1. Diffusion space — where your DTI data lives
2. Structural (T1) space — the high-resolution anatomical image
3. Standard (MNI) space — a common template used for group comparisons

This step computes six transformation matrices that let you move images and coordinates freely between these spaces.

Further reading: FSL Registration Practical — FSL Course covering the two-step diffusion-to-structural-to-standard registration pathway

Conceptual Background

Aligning to a Common Space

Different MRI sequences produce images in different spaces — they have different resolutions, fields of view, and geometric distortions. Registration is needed to:

• Overlay DTI metrics on anatomy: FA maps (in diffusion space) onto T1 images (in structural space)
• Compare across subjects: transform all subjects' DTI data into MNI space for group statistics
• Apply atlas ROIs: atlas parcellations are defined in MNI space; you need to bring them back to each subject's diffusion space for tractography seeding or region-of-interest analysis

Degrees of Freedom (DOF)

FLIRT performs linear registration — it finds the best spatial transformation using a limited set of parameters:

DOF	Name	What It Allows	When to Use
6	Rigid body	3 translations + 3 rotations	Same brain, different contrast (diffusion → structural)
12	Affine	6 rigid + 3 scales + 3 shears	Different brains, similar shape (structural → MNI template)

Diffusion → structural uses 6 DOF because both images are the same person's brain from the same session. The brain did not change shape; it moved slightly between acquisitions and has different contrast, so rigid alignment is enough.

Structural → MNI uses 12 DOF because different people have different brain sizes and shapes, and the affine transformation's scaling and shearing account for those differences.

The Transformation Chain

Diffusion ──(6 DOF)──> Structural ──(12 DOF)──> Standard (MNI)

By concatenating these two transforms, you get a direct diffusion → standard mapping. You also compute the inverse transforms to go in the opposite direction.

Template Selection

Population	Recommended Template	Source
Adults (18–65)	MNI152_T1_2mm_brain	Included with FSL ($FSLDIR/data/standard/)
Older adults (65+)	OASIS template	ANTs templates
Children (4–18)	NIH Pediatric template	NIH Pediatric Database
Infants	Age-specific atlas	dHCP atlas
Study-specific	Build your own	antsMultivariateTemplateConstruction2.sh from ANTs

If your participants are healthy adults, the MNI152_T1_2mm_brain template included with FSL is the standard choice. It is used by most DTI studies and is compatible with all major atlases (JHU, Harvard-Oxford, AAL, etc.).

Prerequisites

Input	Source	Description
FA map	Step 9: DTIFIT	Fractional anisotropy image in diffusion space
Brain-extracted T1	Step 2: Skull Stripping	ANTs skull-stripped structural image
MNI template	FSL installation	$FSLDIR/data/standard/MNI152_T1_2mm_brain

Commands

Step 1: Diffusion → Structural (6 DOF)

# ──────────────────────────────────────────────
# Define paths
# ──────────────────────────────────────────────
dtifit_dir="$base_dir/dtifit/$subj"
ants_dir="$base_dir/ants/$subj"
output_dir="$base_dir/registration/$subj"

mkdir -p "$output_dir"

# ──────────────────────────────────────────────
# Register FA map to skull-stripped T1
# ──────────────────────────────────────────────
flirt -in "$dtifit_dir/${subj}_DTI_FA" \
      -ref "$ants_dir/${subj}_BrainExtractionBrain" \
      -out "$output_dir/${subj}_diff2str" \
      -omat "$output_dir/${subj}_diff2str.mat" \
      -dof 6

Step 2: Structural → Standard (12 DOF)

flirt -in "$ants_dir/${subj}_BrainExtractionBrain" \
      -ref "$FSLDIR/data/standard/MNI152_T1_2mm_brain" \
      -out "$output_dir/${subj}_str2standard" \
      -omat "$output_dir/${subj}_str2standard.mat" \
      -dof 12

Step 3: Compute Inverse Transforms

convert_xfm -omat "$output_dir/${subj}_str2diff.mat" \
            -inverse "$output_dir/${subj}_diff2str.mat"

convert_xfm -omat "$output_dir/${subj}_standard2str.mat" \
            -inverse "$output_dir/${subj}_str2standard.mat"

Step 4: Concatenate Diffusion → Standard

convert_xfm -omat "$output_dir/${subj}_diff2standard.mat" \
            -concat "$output_dir/${subj}_str2standard.mat" \
                    "$output_dir/${subj}_diff2str.mat"

The -concat flag applies transforms in right-to-left order. So -concat A B means "apply B first, then A." In this case: diffusion→structural (B) then structural→standard (A) = diffusion→standard.

Step 5: Inverse of Concatenated

convert_xfm -omat "$output_dir/${subj}_standard2diff.mat" \
            -inverse "$output_dir/${subj}_diff2standard.mat"

Batch Processing Script

#!/bin/bash
# registration.sh — Compute all registration transforms for all subjects

base_dir="/path/to/project"
standard="$FSLDIR/data/standard/MNI152_T1_2mm_brain"

subjects=$(ls -d "$base_dir/dtifit"/sub-* 2>/dev/null | xargs -n1 basename)

for subj in $subjects; do
    echo "Processing: $subj"

    dtifit_dir="$base_dir/dtifit/$subj"
    ants_dir="$base_dir/ants/$subj"
    output_dir="$base_dir/registration/$subj"
    mkdir -p "$output_dir"

    flirt -in "$dtifit_dir/${subj}_DTI_FA" \
          -ref "$ants_dir/${subj}_BrainExtractionBrain" \
          -out "$output_dir/${subj}_diff2str" \
          -omat "$output_dir/${subj}_diff2str.mat" -dof 6

    flirt -in "$ants_dir/${subj}_BrainExtractionBrain" \
          -ref "$standard" \
          -out "$output_dir/${subj}_str2standard" \
          -omat "$output_dir/${subj}_str2standard.mat" -dof 12

    convert_xfm -omat "$output_dir/${subj}_str2diff.mat" \
                -inverse "$output_dir/${subj}_diff2str.mat"
    convert_xfm -omat "$output_dir/${subj}_standard2str.mat" \
                -inverse "$output_dir/${subj}_str2standard.mat"
    convert_xfm -omat "$output_dir/${subj}_diff2standard.mat" \
                -concat "$output_dir/${subj}_str2standard.mat" \
                        "$output_dir/${subj}_diff2str.mat"
    convert_xfm -omat "$output_dir/${subj}_standard2diff.mat" \
                -inverse "$output_dir/${subj}_diff2standard.mat"

    echo "  Done: $subj"
done
echo "Registration complete."

Expected Output

Six transformation matrices per subject:

File	Direction	Purpose
diff2str.mat	Diffusion → Structural	Overlay DTI on anatomy
str2diff.mat	Structural → Diffusion	Bring structural ROIs to diffusion space
str2standard.mat	Structural → MNI	Normalize structural to template
standard2str.mat	MNI → Structural	Bring atlas labels to subject space
diff2standard.mat	Diffusion → MNI	Group analysis in standard space
standard2diff.mat	MNI → Diffusion	Bring atlas ROIs to diffusion space

Quality Check

Apply the diffusion→standard transform and visually verify alignment:

flirt -in "$dtifit_dir/${subj}_DTI_FA" \
      -ref "$FSLDIR/data/standard/MNI152_T1_2mm_brain" \
      -applyxfm -init "$output_dir/${subj}_diff2standard.mat" \
      -out "$output_dir/${subj}_FA_in_MNI"

fsleyes "$FSLDIR/data/standard/MNI152_T1_2mm_brain" \
        "$output_dir/${subj}_FA_in_MNI" -cm red-yellow -a 50 &

Good: FA map aligns with the template — major tracts overlap with expected anatomy. Bad: Obvious misalignment, brain shifted or rotated.

When FLIRT Is Not Enough

FLIRT performs linear registration. For analyses requiring more precise alignment (TBSS, VBA), consider FNIRT (FSL's nonlinear registration) which deforms the image to match the template more precisely.

Common Issues

Issue	Cause	Solution
Poor diff→str alignment	FA map has low contrast	Check DTIFIT output quality
Poor str→standard alignment	Template mismatch for population	Use age-appropriate template
FA in MNI looks stretched	Normal for affine registration	Slight differences expected; use FNIRT if needed

References

• Jenkinson M, Smith SM (2001). A global optimisation method for robust affine registration of brain images. Medical Image Analysis, 5(2), 143-156.
• FSL FLIRT: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FLIRT

Next Step

Proceed to Step 11: Response Function Estimation to begin the CSD modeling chain.

If you also want intracranial volume as a covariate, ICV Calculation is optional and can be run at any point after Step 2.