---

name: registration-transforms description: "Configure SimpleITK transforms, resampling, ImageRegistrationMethod workflows, and optional elastix/transformix wrappers." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

SimpleITK Registration and Transforms

Use this sub-skill when a task involves registration, ImageRegistrationMethod, image-to-image metrics, optimizers, transform initialization, Resample, displacement fields, CompositeTransform, or optional ElastixImageFilter/TransformixImageFilter wrappers.

Route by Task

• For transform direction, physical-space semantics, resampling grids, interpolation, inverse-transform mistakes, and label-vs-intensity interpolation, read transform-resampling.
• For ImageRegistrationMethod setup, initialization, metrics, optimizers, multi-resolution, reproducibility, callbacks, and output-transform use, read registration-workflows.
• For optional ElastixImageFilter, TransformixImageFilter, parameter maps, and build-dependent availability checks, read elastix-transformix.
• For black resamples, no-overlap failures, bad masks/sampling, nondeterminism, type mismatches, and missing optional wrappers, read troubleshooting.
• To validate a local install with a deterministic generated-image translation registration and resample, run registration_smoke.py.

Boundary Notes

• Use ../io-and-data/SKILL.md for ReadImage, WriteImage, transform file IO, ImageIO backend discovery, DICOM, and metadata tags.
• Use ../filtering-segmentation/SKILL.md for preprocessing filters, masks created by thresholding/morphology, segmentation, and label statistics.
• Use ../builds-and-wrapping/SKILL.md for building SimpleITK, Python wrapping, and enabling optional elastix/transformix wrappers.
• Use ../image-core/SKILL.md for image dimensions, spacing/origin/direction, physical coordinates, pixel IDs, and NumPy conversion fundamentals.

Essential Rules

• Import the public package as import SimpleITK as sitk; the Python distribution name is simpleitk.
• Registration output transforms map physical points from the fixed image domain to the moving image domain; pass that transform directly when resampling the moving image onto the fixed image grid.
• Treat images and transforms in physical space: spacing, origin, direction, transform center, and fixed/moving image roles are part of correctness.
• Use sitk.sitkLinear for continuous intensity images and sitk.sitkNearestNeighbor for labels; linear interpolation can invent invalid label values.
• For reproducible sampled registration, pass a fixed seed to SetMetricSamplingPercentage and consider forcing one global thread during smoke tests or exact comparisons.
• Guard every elastix/transformix path with hasattr(sitk, "ElastixImageFilter") and hasattr(sitk, "TransformixImageFilter"); wrappers are optional and were absent from the inspected wheel even though source wrappers exist.