Skip to content

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

BibTeX
@article{konz2026frd,
    title     = {Fr\'{e}chet Radiomic Distance (FRD): A Versatile Metric for
                 Comparing Medical Imaging Datasets},
    author    = {Konz, Nicholas and Osuala, Richard and Verma, Preeti and
                 Chen, Yuwen and Gu, Hanxue and Dong, Haoyu and Chen, Yaqian
                 and Marshall, Andrew and Garrucho, Lidia and Kushibar, Kaisar
                 and Lang, Daniel M. and Kim, Gene S. and Grimm, Lars J. and
                 Lewin, John M. and Duncan, James S. and Schnabel, Julia A. and
                 Diaz, Oliver and Lekadir, Karim and Mazurowski, Maciej A.},
    journal   = {Medical Image Analysis},
    volume    = {110},
    pages     = {103943},
    year      = {2026},
    publisher = {Elsevier},
    doi       = {10.1016/j.media.2026.103943},
    url       = {https://www.sciencedirect.com/science/article/pii/S1361841526000125},
}

[Unreleased]

[1.0.0] — 2026-02-21

Added

  • FRDv1 support — extended feature extraction with ~464 features (Original + LoG + Wavelet), z-score normalisation, and log-transformed Fréchet distance. v1 is now the default.
  • Unified API — compute_frd() supports both v0 and v1 via the frd_version parameter.
  • OOD detection — detect_ood() function and ood CLI subcommand for image-level and dataset-level out-of-distribution detection.
  • Interpretability — interpret_frd() with t-SNE visualisations and per-feature difference rankings.
  • Normalisation reference modes — norm_ref parameter with "joint", "d1", and "independent" options.
  • Paper log transform — use_paper_log flag for Eq. 3 consistency.
  • PyRadiomics extraction controls — bin_width, normalize_scale, voxel_array_shift, log_sigma, config_path, settings_dict.
  • Feature exclusion — exclude_features for post-extraction ablation ("textural", "wavelet", "firstorder", "shape"). The "shape" option removes shape_* and shape2D_* features, which are often constant when no mask is used.
  • Sample matching — match_sample_count to equalise dataset sizes.
  • Means-only mode — means_only for small datasets.
  • MkDocs documentation — hosted on GitHub Pages with full API reference.
  • GitHub Actions CI — automated testing and docs deployment.
  • --version flag — print the installed frd-score version and exit.
  • Pyradiomics compatibility — import guard with helpful error message for broken PyPI release (#903).
  • 3D medical image test fixtures — 8 downsampled (32×32×16) NIfTI volumes from breast MRI in tests/data/medical_3d/{d1,d2}/ (~77 KB total), providing realistic 3D end-to-end test coverage.
  • 2D medical image test fixtures — 20 downscaled (128×128) grayscale PNGs from diverse modalities in tests/data/medical_2d/{d1,d2}/ (~155 KB total), providing realistic end-to-end test coverage.
  • 93+ tests covering v0/v1 computation, normalisation modes, feature exclusion, OOD detection, interpretability, CLI, and real medical image pipelines.

Changed

  • Python requirement bumped from ≥3.8 to ≥3.10 (pyradiomics compatibility).
  • pyradiomics dependency moved to documented install step (PyPI does not support git URLs in install_requires).
  • Default FRD version changed from v0 to v1.
  • pyproject.toml now declares build system.
  • Development status upgraded to Production/Stable.

Fixed

  • Pyradiomics import failure on Python ≥3.10 handled with clear error message.

[0.0.2] — 2024-06-21

Added

  • Initial FRDv0 implementation.
  • CLI via python -m frd_score.
  • Min-max normalisation with FID-compatible range.
  • Save/load .npz statistics.
  • Multi-worker feature extraction.
  • Mask support for localised feature extraction.