github nipreps/mriqc 26.0.0

2 hours ago
  • Docker: docker pull nipreps/mriqc:26.0.0 (linux/amd64 and linux/arm64)
  • PyPI: pip install mriqc==26.0.0 (Python 3.13 or newer)

The first release of the 26.x series, and the first since 24.0.2 (25.0.0rc0 was never finalized). MRIQC now requires Python 3.13 or newer, following SPEC 0, and its container image runs on linux/amd64 and linux/arm64.

With thanks to @asaraog and @vferat for their first contributions, and to @yohanchatelain for proposing --random-seed (#1413).

Highlights

  • Python 3.13 and 3.14, with the SPEC 0 minimum versions of NumPy (2.2), SciPy (1.15), Pandas (2.3), Matplotlib (3.10) and scikit-learn (1.6).
  • A new container image for linux/amd64 and linux/arm64, built with pixi from a lockfile on Python 3.13. AFNI (26.2.05), ANTs (2.6.5) and svht_denoise are compiled from source for both architectures.
  • Reproducible runs: --random-seed makes two runs give identical results.
  • Parquet IQM outputs: each participant-level run writes <entities>_<suffix>+iqms.parquet, with a JSON sidecar holding the schema, the BIDS metadata and the provenance.

Backward-incompatible changes

  • The participant-level <entities>_<suffix>.json file is no longer written: the parquet file and its +iqms.json sidecar replace it.
  • DWI denoising uses svht_denoise instead of MRtrix3's dwidenoise, which is licensed for non-commercial use only and has no arm64 build. DWI IQMs change.
  • Python 3.10, 3.11 and 3.12 are no longer supported.
  • IQMs computed on arm64 differ slightly from those computed on amd64 (most T1w IQMs agree within 1%; BOLD IQMs within 0.02%). Do not mix the two architectures within a study.

Changes

Don't miss a new mriqc release

NewReleases is sending notifications on new releases.