Nipype baseline workflow

[oesteban]

Objective

Build a simple anatomical preprocessing workflow entirely in Nipype, chaining the core anatomical steps — brain extraction, intensity-inhomogeneity (INU) correction, brain-tissue segmentation, and spatial normalization — and run it on the anatomical (T1w) images of a subset of the OpenNeuro testbench (nipreps/fmriprep-equivalence#3). This is the reference implementation the engine-skill milestones (nipreps/agentic-files#14–#20) validate against via Bland–Altman reliability analyses.

Rationale

Each engine-skill milestone reconstructs the same simple anatomical workflow on a different execution target (Claude Code /workflow, bash, CWL, Nextflow, …) and must be shown numerically reliable against a trusted baseline. A plain Nipype implementation of that workflow is the natural baseline: Nipype is the reference engine the whole project ablates against, and a simple anatomical chain is well-trodden — there is an abundance of tutorials and examples to derive it from. Establishing it early, with a fixed testbench execution, gives every downstream engine skill a stable target to be compared to. It has no upstream dependencies and can start immediately.

Methodological approach

1. Assemble the workflow. Chain brain extraction, INU correction, tissue segmentation, and spatial normalization as a standard Nipype workflow over established interfaces (ANTs / FSL / AFNI / FreeSurfer as appropriate), derived from existing tutorials/examples.
2. Run on the testbench T1w subset. Execute on the anatomical (T1w) images of a subset of the selected OpenNeuro testbench datasets (Assemble the fMRIPrep testbench dataset drawing from OpenNeuro fmriprep-equivalence#3, in the making), producing the reference derivatives — brain masks, bias-corrected images, tissue segmentations, and normalized outputs.
3. Publish as the reference. Version-pin inputs and outputs so the derivatives serve as the fixed baseline for the Bland–Altman reliability comparisons in the engine-skill milestones.

Outcomes/impact

1. A simple Nipype anatomical preprocessing workflow (brain extraction → INU correction → tissue segmentation → spatial normalization).
2. Reference derivatives from executing it on the testbench T1w subset.

Impact. This is the trusted baseline every engine-skill milestone (nipreps/agentic-files#14–#20) is measured against. A well-scoped, low-risk task for @effigies, @mgxd, and @oesteban.

Direct upstream issues

(issues this one depends on / builds upon)

None — derivable from existing Nipype tutorials/examples; no board dependency.

Direct downstream issues

(issues that depend on this one)

• Engine skill — Claude Code /workflow agentic-files#14
• Engine skill — bash agentic-files#15
• Engine skill — CWL agentic-files#16
• Engine skill — Nextflow agentic-files#17
• (optional) Engine skill — Pydra agentic-files#18
• Engine skill — Dask agentic-files#19
• Engine skill — Snakemake agentic-files#20