S9: Primary Functional Derivatives

Step Code: S9_primary_functional_derivatives Depends on: S7 (PAM50 warps and spinal levels) Required by: S10 (QC Aggregation & Release)

Purpose

S9 produces the final, analysis-ready functional outputs: the preprocessed BOLD series in both native and PAM50 template space, plus temporal-SNR maps and a per-vertebral-level tSNR table. It performs the cord-aware spatial smoothing and packages everything so an analyst can drop the data straight into a general linear model (GLM).

S9 does not high-pass filter the BOLD (the cosine basis from S8 handles drift) and does not regress out confounds (that is the analyst's GLM, using the S8 table).

Algorithm

1. Cord-aware smoothing

Brain-style isotropic smoothing would blur signal out of the thin cord. Instead, S9 smooths along the straightened cord axis using SCT's sct_smooth_spinalcord (CoSpine lineage), which straightens the cord per volume before applying the kernel.

• Kernel: σ = 1, 1, 5 mm in the right-left, anterior-posterior and superior-inferior directions (FWHM ≈ 2.35 × 2.35 × 11.8 mm). The heavy superior-inferior smoothing exploits signal repetition along the cord while the light in-plane smoothing preserves the cord cross-section.
• An alternative in-plane Gaussian method is available; it applies no through-slice blur.

Both a smoothed and an unsmoothed series are kept, so analysts who prefer to smooth at modelling time can use the unsmoothed data.

2. PAM50 template-space outputs

The BOLD is warped into the PAM50 spinal-cord template. Warping the full 0.5 mm template grid would cost ~17 GB per run; instead S9 crops the template to each run's cord field of view (with padding) and warps directly into that cropped grid, yielding ~1–2 GB per run at the same 0.5 mm resolution. The template-space BOLD ships co-gridded with its cord mask, so it is GLM-ready out of the box. Both smoothed and unsmoothed PAM50 series are emitted, alongside a PAM50 functional reference (temporal mean) and a PAM50 tSNR map.

3. Per-level tSNR

Using S7's PAM50 spinal-level labels, S9 writes a table of mean and standard- deviation tSNR per vertebral level — the cohort-comparison metric S10 plots as a heatmap.

Step Metric and QC

The step-local truth metric is the in-cord tSNR ratio (post-smoothing / pre-smoothing). Cord-aware smoothing should raise tSNR by roughly 1.5–2.5×.

S9 also verifies the achieved smoothness by estimating per-axis FWHM in the cord mask and checking it against the requested kernel within tolerance.

Outputs

derivatives/spineprep/{dataset}/sub-{id}/func/
├── sub-{id}_..._space-PAM50_desc-smoothed_bold.nii.gz     # GLM-ready, template space
├── sub-{id}_..._space-PAM50_desc-unsmoothed_bold.nii.gz
├── sub-{id}_..._space-PAM50_desc-funcref.nii.gz           # temporal mean
├── sub-{id}_..._space-PAM50_desc-tsnr.nii.gz              # tSNR map
├── sub-{id}_..._space-PAM50_desc-cordmask.nii.gz          # co-gridded cord mask
└── sub-{id}_..._desc-tsnr_per_level.tsv                   # per-vertebral-level tSNR

derivatives/spineprep/{dataset}/sub-{id}/figures/
└── sub-{id}_..._desc-S9_tsnr_map_axial.png

References

1. Cord-aware smoothing: Spinal Cord Toolbox sct_smooth_spinalcord; De Leener et al. NeuroImage 145:24–43 (2017). DOI
2. PAM50 template: De Leener et al. NeuroImage 165:170–179 (2018). DOI

Last updated: June 2026