v0.1.0

Initial research release of Polygonal Path Image, by Paula Agregán Reboredo, Vincent Bismuth, and Laurent Najman.

• Python/Cython API for minimum-cost polygonal paths, voting, tortuosity filtering, orientation, and pruning.
• Repairs historical path reconstruction while preserving minimum costs and deterministic ties.
• Full-image validation checks 209,921 paths with no reconstructed-cost or geometry inconsistencies in the package. The supplied historical images contained 7,128 and 26,795 inconsistent legacy reconstructions.
• 202 tests cover an independent exhaustive small-image oracle, input validation, postprocessing, and the independent research audit.
• 20 prebuilt wheels cover CPython 3.10–3.14, Linux x86-64 (glibc 2.28+), Windows x86-64, and macOS Intel/Apple Silicon. Every wheel is tested natively with NumPy 2; Python 3.10–3.12 wheels also run the suite with NumPy 1.26.4.
• BSD-3-Clause license, software credits, and the MICCAI 2012 paper citation are included.

Install from PyPI:

python -m pip install polygonal-path-image==0.1.0

Download the wheel matching your Python and platform and install its actual filename with python -m pip install ./<downloaded-wheel>.whl. These wheels contain the compiled extension. The source archive requires a C compiler. SHA256SUMS covers all 21 distribution files.

This release implements the later four-cone discrete student variant, with documented raster geometry, source-cost, and postprocessing conventions. Validation establishes numerical correctness and historical compatibility on the tested cases; it does not reproduce the paper's clinical evaluation.

Read the validation report, API, and release/platform details. The reference paper is Bismuth et al., MICCAI 2012.