---

name: image-ops-math description: "Use ANTsPy image operations for smoothing, resampling, masking, cropping, morphology, iMath, bias correction, histogram matching, and image metrics." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Image Operations and Math

Use this sub-skill when a task asks for ANTsPy preprocessing, filtering, masks, crop/pad/slice, resampling, intensity normalization, morphology, iMath, similarity metrics, neighborhood extraction, or quantitative image utilities.

Route First

• Use image-core for image creation, IO, cloning, pixel types, metadata checks, NumPy conversion, and physical-space consistency helpers before applying operations here.
• Use registration-transforms for registration, non-identity transform application, transform ordering, point transforms, and transform output troubleshooting.
• Use segmentation-labels for label statistics, label geometry, label overlap, segmentation workflows, and multi-label morphology beyond generic binary/grayscale operations.
• Use visualization-interop for plotting, screenshots, notebook display, and conversions whose purpose is visualization.

Read These References

• API reference: verified public signatures, parameter choices, chainable methods, and scalar/vector/RGB caveats.
• Workflows: preprocessing, masks, resampling, bias correction, morphology, metrics, neighborhood, and validation recipes.
• iMath reference: ANTsPy-recognized iMath operations, helper wrappers, operation families, and unsupported-operation recovery.
• Troubleshooting: interpolator mistakes, physical-vs-voxel resampling, mask dimensionality, iMath failures, backend errors, and expensive 3D operations.
• Smoke script: tiny in-memory check covering threshold, smoothing, resampling, masking, iMath, morphology, metrics, and expected error recovery.

Default Patterns

• Prefer ants.resample_image_to_target(image, target, interp_type="linear") when output must inherit another image's shape, spacing, origin, and direction; use "nearestNeighbor" or "genericLabel" for label images.
• Use ants.resample_image(image, params, use_voxels=True, interp_type=0) when params are output voxel counts; use use_voxels=False when params are new physical spacing values.
• Build tiny or synthetic masks with ants.get_mask(image, low_thresh=..., high_thresh=..., cleanup=0) because default cleanup can erase small objects.
• Use ants.iMath(image, "Normalize"), ants.iMath(image, "TruncateIntensity", low, high, bins), or helper wrappers such as ants.iMath_normalize(image) for common operations.
• Split long chained ANTsImage calls when a step changes shape or physical space so you can inspect dimension, shape, spacing, origin, and direction before combining images.