Registration

Overview

This workflow registers a postmortem ex vivo T2w hemisphere to its matched antemortem in vivo T1w MRI. Registration is driven primarily by label maps (GM and WM-plus masks), and the resulting transforms are used to warp ex vivo MRI intensities and segmentations into in vivo space.

The implementation is provided as a bash script:

scripts/registration_exvivo_invivo_greedy_v2.sh

Key Ideas

  • In vivo segmentations are obtained using SynthSeg on 1mm-conformed T1w.

  • Ex vivo MRI and 10-label segmentations are resampled to 1mm for compatibility.

  • Labels are harmonized across ex vivo and in vivo to create a common registration scheme.

  • Registration proceeds in stages using greedy:

    1. Moments-based initialization

    2. Affine (12 DOF)

    3. Deformable

  • Final outputs include ex vivo MRI/segmentations warped into the in vivo hemisphere space.

Requirements

  • greedy binaries available in PATH

  • c3d (Convert3D)

  • FreeSurfer utilities (used by the script), e.g., mri_label2vol

  • SynthSeg output for in vivo T1w (1mm space)

Inputs

You will need directories for:

  • Ex vivo MRI (T2w) volumes

  • Ex vivo 10-label segmentations (purple-mri output)

  • In vivo MRI (T1w) volumes (QCed + resampled/conformed to 1mm)

  • In vivo SynthSeg segmentations (derived from the 1mm T1w)

The script expects subject naming conventions consistent with your dataset (INDD IDs) and uses a subjects=(...) list to define which cases to process.

Processing Steps (What the Script Does)

1) Clean the ex vivo 10-label segmentation (largest connected component per label)

For each label in the ex vivo segmentation, the script retains the primary connected component to remove small spurious islands, then merges labels back into a cleaned segmentation.

2) Resample ex vivo MRI and segmentation to 1mm

The ex vivo MRI is resampled to 1mm isotropic, and the segmentation is resampled into the MRI grid using FreeSurfer mri_label2vol with header-based alignment.

3) Harmonize in vivo SynthSeg labels to a compact hemisphere label set

SynthSeg labels are retained and remapped into a simplified label scheme. The mapping differs for right vs left hemispheres (the script auto-detects hemisphere based on file existence).

4) Harmonize ex vivo labels to the same scheme

The ex vivo 10-label map is remapped to match the simplified in vivo scheme used for registration.

5) Build GM and WM-plus binary masks

Two masks are constructed for robust registration:

  • GM mask (typically cortical GM plus selected GM class)

  • WM-plus mask (WM and deep structures grouped)

These are extracted for both ex vivo and in vivo, after trimming the in vivo hemisphere volume.

6) Run greedy registration (moments → affine → deformable)

The script performs:

  • Moments-based alignment (initialization)

  • 12-DOF affine alignment initialized from moments

  • Deformable alignment initialized from affine

Registration uses GM and WM-plus masks as paired inputs.

7) Warp ex vivo MRI and segmentation into in vivo space

Using the estimated transforms, the script produces:

  • ex vivo MRI warped to in vivo space (linear interpolation)

  • ex vivo segmentation warped to in vivo space (label interpolation)

How to Run

  1. Open:

scripts/registration_exvivo_invivo_greedy_v2.sh

  1. Set paths at the top of the script:

  • exvivo_mri_dir

  • exvivo_segm_dir

  • exvivo_segm_cleaned_dir

  • invivo_dir_mri

  • invivo_dir_segm

  • work_dir

  1. Set the subject list:

subjects=(INDD_XXXXXX_count_1_acq_...  INDD_YYYYYY_count_2_acq_...)

  1. Run:

bash scripts/registration_exvivo_invivo_greedy_v2.sh

Outputs

Per subject, the script typically writes:

  • Cleaned ex vivo segmentation (connected-component filtered)

  • 1mm-resampled ex vivo MRI and segmentation

  • In vivo simplified segmentation (hemisphere relabeled)

  • Greedy transforms:

    • moments.mat

    • affine.mat

    • warp.nii.gz

  • Warped ex vivo products in in vivo space:

    • ex vivo MRI registered (moments / affine / deformable)

    • ex vivo segmentation registered (moments / affine / deformable)

Notes

  • This script is opinionated to a particular file naming convention and directory layout. Please read the header comments and adjust variable names and file patterns as needed.

  • The workflow assumes SynthSeg is run in 1mm space for the in vivo T1w.

  • Label mappings differ across hemispheres; the script handles R/L by checking which ex vivo file exists.

Next Step

Proceed to:

Group Analysis

to run ROI-wise and/or voxel/vertex-wise statistical analyses once subjects are aligned.