Change interpolation defaults to "linear"

[LMZimmer]

Changed defaults for niftireg and ANTS backends to linear which was NN interpolation before.

An "interpolator" argument is now available upon Registrator construction and .register / .transform calls.

[LMZimmer]

Adresses Issue #192.

[Copilot]

Copilot review overview

🟡 Changes recommended

Defacing masks need label-preserving interpolation, and registration overrides fail with unsupported backends.

Review effort: Balanced
Findings: 1 · 1

Open (2)

• Use nearest-neighbor interpolation when resampling defacing masks · New
• Reject unsupported interpolator options for non-ANTs backends · New

What changed in this PR

Updates image resampling to use linear interpolation by default and exposes interpolation overrides through registration and modality APIs.

Changes:

• Switches ANTs and NiftyReg transformation defaults to linear.
• Adds ANTs interpolation selection from configured defaults or call arguments.
• Forwards modality-level interpolation options and updates documentation.

File	Description
brainles_preprocessing/​transform.py	Documents linear interpolation defaults.
brainles_preprocessing/​registration/​niftyreg/​niftyreg.py	Changes forward and inverse transformation defaults.
brainles_preprocessing/​registration/​ANTs/​ANTs.py	Resolves interpolation settings and forwards registration overrides.
brainles_preprocessing/​modality.py	Adds interpolation arguments to registration and transformation wrappers.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[neuronflow]

@LMZimmer please have a look at copilot comments. The masking needs to haben with nearest neighbor interpolation

[neuronflow]

@LMZimmer please see comment above

[Copilot]

🔵 Needs a closer look

The new modality-level interpolation option can silently produce output using an unintended interpolation method.

0 open findings

2 resolved since last review

• Use nearest-neighbor interpolation when resampling defacing masks
• Reject unsupported interpolator options for non-ANTs backends

Previously missed (1)

In code that hasn't changed since last review

Elastix ignores the requested interpolation option

brainles_preprocessing/​modality.py:404

The new Modality.transform(interpolator=...) option is silently ignored for ElastixRegistrator. Its transform() accepts the keyword but never uses it when constructing the parameter object or calling register() (registration/elastix/elastix.py:83-111), so the output uses the backend default instead of the requested interpolation. Reject explicit interpolation options for Elastix here, as Modality.register() does for unsupported backends, or implement support in the backend.

🧠 Review effort: Balanced

---

Give feedback about Copilot approvals in this survey to enter a drawing for a $150 gift card.