Fix quickshear edge detection wrap-around issue by using convolution

[uturkbey]

Bug fix for the edge detection wrap-around issue mentioned in #167:

• np.roll operations are changed to scipy.ndimage.convolve with a 4-neighborhood Laplacian edge detection kernel.
• Tested over various real and synthetic brain masks, and outputs are found to be identical to baseline for brain-not-touching-edge cases.
• SciPY already in project dependencies.
• Runtime and peak memory usage are benchmarked against baseline and another solution (np.pad before np.roll). Summarized results are as below.
  

[uturkbey]

Please also see #167 (comment)

[neuronflow]

@uturkbey thanks! Please try to fix the failing tests. Unfortunately, you cannot use our brainless bot for black formatting because the PR does not belong to the repo.

[Copilot]

Pull request overview

This pull request fixes a critical bug in the Quickshear defacing algorithm where edge detection produced incorrect results when brain masks touch array boundaries. The issue was caused by np.roll operations creating artificial wrap-around edges on opposite boundaries.

Changes:

• Replaced np.roll-based edge detection with scipy.ndimage.convolve using a 4-neighborhood Laplacian kernel
• Added import for convolve from scipy.ndimage
• Applied mode='constant', cval=0.0 to prevent boundary wrap-around

---

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

[neuronflow]

@uturkbey @copilot I am wondering whether it would be enough to simply pad with background and crop after the operation instead of the whole convolution?

Shall we have a call and discuss soon?

[uturkbey]

@uturkbey @copilot I am wondering whether it would be enough to simply pad with background and crop after the operation instead of the whole convolution?

Shall we have a call and discuss soon?

For sure, writing you an email!