Pull request overview

Aligns skull-stripping mask application with the defacing workflow by supporting either a global masking/background value or a per-image minimum when filling voxels outside the brain mask.

Changes:

• Added masking_value support to BrainExtractor and updated apply_mask() to fill outside-mask voxels using either the provided global value or the per-image minimum.
• Updated HDBetExtractor and SynthStripExtractor constructors to accept/forward masking_value.
• Minor whitespace cleanup in the registration package init.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Do we need changes here?

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

brainles_preprocessing/brain_extraction/brain_extractor.py:85

• The masking workflow now has two branches (explicit masking_value vs per-image minimum) and changes the masking semantics from multiplication to np.where(..., current_masking_value), but existing tests only assert that an output file is created. Please add assertions covering the new behavior (e.g., that voxels outside the mask equal np.min(input_data) when masking_value=None, and equal the provided value when masking_value is set) to prevent regressions.

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(input_data)
        else:
            current_masking_value = (

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

brainles_preprocessing/brain_extraction/synthstrip.py:39

• masking_value is now accepted and passed into BrainExtractor, but SynthStrip still hard-codes the masked background value inside extract() (see bg = np.min([0, img_data.min()]) and img_data[mask == 0] = bg at lines 225-226). This means center-modality skull-stripped outputs written by SynthStripExtractor.extract() will ignore the new masking_value behavior described in the docstring/PR.

            border (int): Mask border threshold in mm. Defaults to 1.
            masking_value (Optional[Union[int, float]], optional): global value to be inserted in the masked areas. Default is None which leads to the minimum of each respective image.

        """
        super().__init__(masking_value=masking_value)

brainles_preprocessing/brain_extraction/brain_extractor.py:79

• The new masking_value behavior in apply_mask() (global override vs per-image minimum) is not covered by assertions in the existing tests. Current tests only check that an output file is created, so regressions in the chosen background value or dtype casting would go unnoticed.

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(input_data)
        else:
            current_masking_value = (

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

brainles_preprocessing/brain_extraction/brain_extractor.py:6

• Enum is imported but not used in this module. Keeping unused imports can confuse readers and may fail linting if/when a linter is added.

from enum import Enum
import numpy as np

brainles_preprocessing/brain_extraction/brain_extractor.py:85

• New masking behavior (global masking_value vs per-image minimum) is implemented here, but existing tests only assert that an output file is created. Add unit coverage that verifies masked voxels are set to the configured value, and that None uses the input image minimum.

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(input_data)
        else:
            current_masking_value = (
                np.array(self.masking_value).astype(input_data.dtype).item()
            )
        # Apply mask (element-wise either input or masking value)
        masked_data = np.where(
            mask_data.astype(bool), input_data, current_masking_value
        )

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

brainles_preprocessing/brain_extraction/brain_extractor.py:84

• The masking/background value behavior changed from multiplying by the mask to inserting either a per-image minimum or a user-supplied global value, but existing unit tests only assert that an output file is created. This makes regressions (e.g., background incorrectly set) hard to catch. Add assertions in tests to verify: (1) when masking_value=None, voxels outside the mask equal input_data.min(); and (2) when masking_value is provided, voxels outside the mask equal that value (cast to the input dtype).

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(input_data)
        else:
            current_masking_value = (
                np.array(self.masking_value).astype(input_data.dtype).item()
            )
        # Apply mask (element-wise either input or masking value)
        masked_data = np.where(
            mask_data.astype(bool), input_data, current_masking_value
        )

Pull request overview

Copilot reviewed 3 out of 3 changed files in this pull request and generated no new comments.

brainles_preprocessing/brain_extraction/synthstrip.py:229

• The masking-value resolution logic is duplicated here and in BrainExtractor.apply_mask (same None/min vs global value + dtype-cast). Keeping two copies increases the risk of future divergence between the generic masking path and SynthStrip’s masking behavior. Consider extracting this into a shared helper on BrainExtractor (e.g., _resolve_masking_value(input_data: np.ndarray) -> scalar) and reusing it from both places.

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(img_data)
        else:
            current_masking_value = (

brainles_preprocessing/brain_extraction/brain_extractor.py:78

• New behavior selects a per-image minimum when masking_value is None, and otherwise forces the provided masking_value into the image dtype before applying the mask. The existing tests for apply_mask only assert that an output file is created; they don’t verify that the masked voxels actually get the expected fill value (both for the default min case and the explicit global masking_value case). Adding assertions for these cases would protect against regressions (e.g., dtype-cast/rounding or incorrect fallback selection).

        # check whether a global masking value was passed, otherwise choose minimum
        if self.masking_value is None:
            current_masking_value = np.min(input_data)
        else:
            current_masking_value = (