Parcellation

Overview

Once you have obtained a topology-corrected volumetric segmentation, you can proceed to the surface-based pipeline to obtain whole-hemisphere cortical parcellations in standard neuroanatomical atlases.

This step runs locally (CPU only; no GPU required).

The pipeline prepares the data, computes necessary transformations, performs surface modeling and topology stabilization, and generates atlas-based parcellations.

Requirements

  • FreeSurfer installed locally (tested with FreeSurfer 7.4.0 on Linux)

  • Python dependencies listed in dependencies.txt Install using:

    pip install -r dependencies.txt

  • A topology-corrected volumetric segmentation (e.g., 10-label output)

Optional: 1mm Conforming

You may run the pipeline at 1mm resolution by conforming the MRI and segmentation to MNI space.

Reference snippet: https://github.com/Pulkit-Khandelwal/purple-mri/blob/main/misc_scripts/flip_conform.sh

Running the Surface-Based Pipeline

Clone the repository and run run_surface_pipeline.sh from within the purple_mri directory.

Inputs

The script requires the following arguments:

  • freesurfer_path — path to the FreeSurfer installation

  • working_dir — directory where outputs will be stored

  • mri_path — directory containing MRI images (NIfTI)

  • segm_path — directory containing topology-corrected segmentations

  • external_atlases_path — directory containing additional atlas resources

  • num_threads — number of CPU threads

  • hemis — hemisphere flag (rh or lh)

Data Organization

  • Place MRI volumes in mri_path

  • Place corresponding segmentation volumes in segm_path

  • Ensure MRI and segmentation filenames match exactly (both ending in .nii.gz)

  • Place fsaverage in working_dir

Command

cd purple_mri

bash run_surface_pipeline.sh \
  freesurfer_path \
  working_dir \
  mri_path \
  segm_path \
  external_atlases_path \
  num_threads \
  rh

Atlases

The pipeline produces parcellations in commonly used atlases, including:

  • Desikan–Killiany–Tourville (DKT)

  • Schaefer

  • Glasser

  • von Economo–Koskinas

Outputs

Typical outputs include:

  • White and pial cortical surfaces

  • Native-space atlas parcellations

  • ROI-level statistics

  • Vertex-wise cortical measures (e.g., thickness)

Next Step

Proceed to:

Group Analysis

for vertex-wise and ROI-wise statistical modeling.