S6: Registration

Step Code: S6_func_to_anat_registration Depends on: S5 (Distortion Correction), S2 (Anatomical cord reference) Required by: S7 (Normalization), S8 (Confounds)

Purpose

S6 aligns the functional data to the subject's own anatomy. It registers the distortion-corrected mean BOLD from S5 to the anatomical cord reference from S2, and stores the transform in both directions:

• a forward warp (BOLD to anat), and
• an inverse warp (anat to BOLD).

S7 (PAM50 template normalization) and S8 (confound extraction) reuse these warps rather than recomputing them. The registration is intentionally driven by the cord segmentation, not by image intensity, because a cord-cropped EPI has an intensity cost surface dominated by the air surrounding the cord — an intensity-only registration diverges on this geometry.

Algorithm

S6 is a single sct_register_multimodal call with a three-stage, cord-segmentation-driven parameter string. This is the cord-fMRI field-standard recipe (CoSpi 1FOV/2FOV; also used under Kaptan 2023 SCT conventions). The functional image is the moving image and the anatomy is the destination (-i funcref -d anat) — fewer slices on the destination is faster and avoids resampling the BOLD onto the anat grid.

Before registration, the anatomy is cropped to a dilated cord region (sct_crop_image -m anat_seg -dilate 10x10x10), which restricts the cost surface to cord context.

All three stages use type=seg (cord segmentation as the cost), the MeanSquares metric, and slicewise=1. Resampling uses spline interpolation.

Step Metric and QC

The step-local truth metric is 3D cord Dice — the overlap between the EPI cord (warped into anat space) and the anatomy cord segmentation. This is the standard cord-registration validation measure (Cohen-Adad 2014). Three further metrics support it:

• HD95 (95th-percentile Hausdorff distance) — boundary disagreement. Dice can be high while a few cord voxels are far off; HD95 catches that. It is quantized to the EPI slice thickness (≈ 5 mm) and sensitive to single end-slice segmentation dropout, so it is treated as observability-only (it can soft-WARN but never FAILs).
• ASD (average symmetric surface distance) — mean boundary disagreement, less outlier-sensitive than HD95.
• Centerline round-trip drift — how far the cord centerline moves under forward-then-inverse warp. Because bsplinesyn optimizes forward and inverse separately, some drift is intrinsic even at Dice 0.95; this is observability-only, with permissive thresholds.
• mi_after — mutual information between BOLD and anat in registered space; a legacy sanity check, not a gate.

Diagnostic Reportlets

• bold_on_anat — the BOLD shown against the anatomy with the cord contour overlaid, in axial and sagittal views. If the cord contour sits off the cord, Dice is low; if the cord rises or falls relative to the anatomy along Z, the columnwise stage failed.
• cord_dice_per_slice — a per-Z Dice bar chart, color-coded by PASS/WARN/FAIL. A few low-Dice slices point to an HD95 outlier; uniformly middling Dice points to a global mis-registration.

Outputs

derivatives/spineprep/{dataset}/sub-{id}/func/
├── sub-{id}_..._from-bold_to-anat_xfm.nii.gz     # forward warp (+ .json sidecar)
├── sub-{id}_..._from-anat_to-bold_xfm.nii.gz     # inverse warp
├── sub-{id}_..._space-anat_desc-mean_bold.nii.gz # mean BOLD in anat geometry (QC)
└── sub-{id}_..._desc-tsnr_funcref.nii.gz         # tSNR reference (used by S7)

derivatives/spineprep/{dataset}/sub-{id}/figures/
├── sub-{id}_..._desc-S6_bold_on_anat.png
└── sub-{id}_..._desc-S6_cord_dice_per_slice.png

The forward-warp JSON sidecar carries the reproducibility receipt: policy SHA, source path, registration method and parameters, and software versions.

QC Status Logic

status = PASS if:
  - cord Dice >= pass_dice_min (0.85), OR
  - cord Dice >= pass_dice_min_syn_fallback (0.80) when S5 used the SyN fallback
    (a looser bar is realistic — a no-fieldmap BOLD is geometrically less faithful)

status = FAIL if:
  - cord Dice < fail_dice_below (0.65)

status = WARN otherwise (Dice in [0.65, pass gate))

HD95 and the centerline round-trip are reported but do not FAIL a run. On the 11-run validation cohort all runs PASS, with Dice 0.89-0.97 on most datasets and a floor of 0.80 in the SyN-fallback regime.

References

1. SCT registration: De Leener et al. NeuroImage 145:24-43 (2017).
2. Cord registration validation (Dice): Cohen-Adad et al. (2014).
3. Cord resting-state / SCT conventions: Kaptan et al. NeuroImage (2023).
4. CoSpine database: Wei et al. Scientific Data (2025).

Last updated: July 2026