Skip to content

Pipeline Overview

Target Audience

Neuroimaging researchers and pipeline administrators.

This page describes the high-level architecture and processing flow of the BRC Structural Pipeline.

Architecture

The pipeline is orchestrated by a single top-level entry script (struc_preproc.sh) that dispatches work to three sequential processing parts. Each part can be run locally (sequentially) or submitted as an independent cluster job when CLUSTER_MODE=YES.

struc_preproc.sh
  ├── struc_preproc_part_1.sh      # T1 + T2 preprocessing
  │     ├── run_T1_preprocessing.sh
  │     │     └── run_T1_sienax.sh
  │     └── run_T2_preprocessing.sh
  │           └── run_T2_bianca.sh
  ├── struc_preproc_part_2.sh      # FreeSurfer (optional)
  │     └── recon-all
  └── struc_preproc_part_3.sh      # FastSurfer + output organisation
        ├── run_fastsurfer.sh
        └── output_organization.sh

Processing Flow

Part 1 — Preprocessing

All core image processing happens in Part 1. For the T1 image this includes:

  1. Reorientation to standard orientation (fslreorient2std)
  2. Field-of-view cropping (robustfov)
  3. Initial brain extraction (FSL BET, recursive)
  4. FOV reduction (standard_space_roi)
  5. Linear registration to MNI152 1mm (FLIRT, 12 DOF)
  6. Non-linear registration to MNI152 (FNIRT or ANTs SyN, depending on --regtype)
  7. Atlas-based brain mask extraction
  8. De-facing (MNI facemask warped to native space)
  9. Bias field estimation and correction (FSL FAST)
  10. Tissue-type segmentation (CSF / GM / WM partial volume maps)
  11. Subcortical segmentation (FSL FIRST, optional)
  12. Brain volume estimation (SIENAX)

For the T2 image (if provided), Part 1 also:

  1. Co-registers T2 to T1 space
  2. Applies the T1 brain mask and bias field
  3. Computes T2-to-MNI registration
  4. Applies de-facing
  5. Runs BIANCA white matter lesion detection

Part 2 — FreeSurfer (optional)

Runs recon-all -all for full cortical parcellation and surface reconstruction. When a T2-FLAIR image is provided, it is passed to recon-all via the -FLAIR flag for improved cortical surface accuracy.

Part 3 — FastSurfer + Output Organisation

Optionally runs FastSurfer for rapid DNN-based cortical reconstruction. Regardless of the surface option chosen, Part 3 always calls output_organization.sh to move all temporary outputs into the canonical BRC folder hierarchy and optionally runs multi-channel tissue segmentation (FSL FAST joint T1+T2).

Registration Strategies

The pipeline supports three registration modes selected via --regtype:

--regtype Method Notes
1 Linear only (FLIRT 12 DOF) Fastest; brain mask from BET propagated via linear transform
2 (default) Linear + FNIRT non-linear Recommended; brain mask from warped MNI atlas; requires ~12 GB RAM
3 Linear + ANTs SyNQuick Highest accuracy; requires ANTs and c3d; slow without multi-core

Cluster Mode

When CLUSTER_MODE=YES, the three parts are submitted as dependent cluster jobs via jobsub:

  • Job 1 (Part 1): T1+T2 preprocessing. Time limit: 12 h (RegType 2) or 5 h (RegType 1 or 3).
  • Job 2 (Part 2, FreeSurfer): runs in parallel with Job 1. Time limit: 24 h.
  • Job 3 (Part 3): waits on Job 1 (and Job 2 if FreeSurfer enabled). Time limit: 7.5 h (FastSurfer) or 15 min (output organisation only).

ANTs (RegType 3) uses 24 CPU cores in cluster mode; all other modes use 1 core.