Add implementation of Aggregated Jaccard Index (AJI) metric

[Copilot]

This PR implements the Aggregated Jaccard Index (AJI) as described in Kumar et al. (2017) for evaluating instance segmentation performance in biomedical image analysis.

What is AJI?

The Aggregated Jaccard Index is a comprehensive metric that evaluates segmentation quality by:

1. Finding optimal object matches between ground truth and predicted objects using IoU
2. Computing aggregated intersection and union across all matched pairs
3. Including unmatched objects in the denominator to penalize false positives and false negatives

The mathematical formulation is:

AJI = Σ |G_k ∩ P_σ(k)| / (Σ |G_k ∪ P_σ(k)| + Σ |P_j unmatched|)

Implementation Details

New AggregatedJaccardIndex class:

• Inherits from AsymmetricMeasureMixin and Measure following existing patterns
• Configurable parameters:
  • min_ref_size (default: 1) - Skip small ground truth objects
  • iou_threshold (default: 0.5) - Minimum IoU for considering objects as matched
• Proper handling of edge cases (empty images, no matches, etc.)
• Integration with the existing Study framework

Usage example:

import segmetrics as sm
import numpy as np

# Create ground truth and prediction
expected = np.array([[1, 1, 0], [1, 1, 0], [0, 0, 2]], dtype=np.uint8)
actual = np.array([[1, 1, 0], [1, 0, 0], [0, 0, 2]], dtype=np.uint8)

# Compute AJI
aji = sm.AggregatedJaccardIndex()
aji.set_expected(expected)
result = aji.compute(actual)
print(f"AJI: {result[0]}")  # 0.8

Changes Made

1. Added AggregatedJaccardIndex class in segmetrics/regional.py
2. Updated exports in segmetrics/__init__.py to include the new measure
3. Added comprehensive tests covering various scenarios:
   • Perfect matches and partial overlaps
   • Empty images and no-match cases
   • IoU threshold behavior
   • Complex multi-object scenarios
4. Updated test suite to include AJI in default name tests and study integration

Validation

The implementation has been thoroughly tested with:

• Unit tests covering edge cases and parameter validation
• Integration tests with the Study framework
• Manual validation against the Kumar et al. (2017) specification
• Comparison with expected mathematical results

This addition provides researchers in biomedical image analysis with a standardized implementation of this important metric for evaluating instance segmentation algorithms.

Reference

Kumar, N., Verma, R., Sharma, S., Bhargava, S., Vahadane, A., & Sethi, A. (2017). A dataset and a technique for generalized nuclear segmentation for computational pathology. IEEE Transactions on Medical Imaging, 36(7), 1550-1560.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[kostrykin]

@copilot Instead of adding the test_aggregated_jaccard_index method to the MeasureTest test case class, create a new test case class called AJITest add the tests to this one.

[Copilot]

@copilot Instead of adding the test_aggregated_jaccard_index method to the MeasureTest test case class, create a new test case class called AJITest add the tests to this one.

Done! I've moved the test_aggregated_jaccard_index method from the MeasureTest class to a new dedicated AJITest class as requested. The tests are now properly organized and continue to pass. Commit: a62106f

[kostrykin]

@copilot Split the different test cases in AJITest.test_aggregated_jaccard_index ("Test case 1", "Test case 2", etc.) into separate test methods of the test case class, like .test_perfect_match, test_partial_overlap, etc.

[Copilot]

@copilot Split the different test cases in AJITest.test_aggregated_jaccard_index ("Test case 1", "Test case 2", etc.) into separate test methods of the test case class, like .test_perfect_match, `...

Done! I've split the test_aggregated_jaccard_index method into 7 separate test methods with descriptive names:

• test_perfect_match
• test_partial_overlap
• test_no_overlap
• test_empty_images
• test_ground_truth_empty_prediction_non_empty
• test_prediction_empty_ground_truth_non_empty
• test_iou_threshold

Each test method now has a clear, focused purpose and all tests continue to pass. Commit: 966e5e3

[kostrykin]

@copilot Fix flake8 linting issues in segmetrics/regional.py

[Copilot]

@copilot Fix flake8 linting issues in segmetrics/regional.py

Fixed all flake8 linting issues in segmetrics/regional.py. The issues addressed include:

• Fixed E501 line too long errors by breaking long lines appropriately
• Removed W293 blank line whitespace errors
• Fixed W291 trailing whitespace
• Corrected F631 assertion syntax error

All tests pass and AJI functionality remains intact. Commit: 345a75f

[github-actions]

Test coverage of 9ffee05: 88%