parcellate 0.4.1

Extract regional statistics from scalar brain maps using atlas-based parcellation.

parcellate

Extract regional statistics from scalar neuroimaging maps using atlas-based parcellation.

What It Does

parcellate is a Python library that extracts regional statistics from volumetric brain images using atlas-based parcellation. Given a scalar map (e.g., gray matter density, fractional anisotropy) and a labeled atlas, it computes summary statistics for each brain region and returns a tidy pandas.DataFrame.

Key Features:

• 45 built-in statistics in three tiers (core / extended / diagnostic), including robust estimates, higher-order moments, and normality tests
• Flexible atlases: discrete 3D integer-label atlases and 4D probabilistic atlases (e.g., XTRACT tracts)
• Built-in masks: MNI152 grey-matter, white-matter, and brain masks via nilearn
• Custom statistics: supply any callable as a statistic
• Scikit-learn–style API: fit / transform pattern with smart resampling caching

Installation

pip install parcellate

Or from a local checkout:

git clone https://github.com/GalKepler/parcellate.git
cd parcellate
pip install -e .

Quick Start

from parcellate import VolumetricParcellator

parcellator = VolumetricParcellator(
    atlas_img="atlas.nii.gz",
    lut="atlas_lut.tsv",       # TSV with "index" and "label" columns
    mask="gm",                 # built-in MNI152 grey-matter mask
    stat_tier="extended",      # 21 statistics (default: all 45)
)

parcellator.fit("subject_map.nii.gz")
stats = parcellator.transform("subject_map.nii.gz")
print(stats.head())

Statistics Tiers

Select a tier with stat_tier to control the number of computed statistics:

Tier	Count	Use case
core	6	Fast exploration, large cohorts
extended	21	Production pipelines
diagnostic (default)	45	QC, distribution inspection

Core: mean, std, median, volume_mm3, voxel_count, sum

Extended adds: robust means (MAD, z-score, IQR filtered), dispersion (cv, robust_cv), shape (skewness, excess_kurtosis), percentiles (5th, 25th, 75th, 95th)

Diagnostic adds: normality tests (Shapiro-Wilk, D'Agostino K²), outlier proportions, tail mass, entropy, boolean QC flags

See the full Metrics reference for descriptions of all 45 statistics.

Custom Statistics

import numpy as np
from parcellate import VolumetricParcellator

parcellator = VolumetricParcellator(
    atlas_img="atlas.nii.gz",
    stat_functions={
        "iqr": lambda x: float(np.nanpercentile(x, 75) - np.nanpercentile(x, 25)),
        "q90": lambda x: float(np.nanpercentile(x, 90)),
    },
)

Output

transform() returns a pandas.DataFrame with one row per atlas region:

Column	Description
index	Integer region index from the atlas
label	Region name from the lookup table
mean, std, …	Computed statistics (depends on stat_tier)

Development

make install   # create venv + install pre-commit hooks
make test      # run tests with coverage
make check     # lock file, pre-commit, deptry

Documentation

Full documentation: https://GalKepler.github.io/parcellate/

License

MIT License — see the LICENSE file for details.

Citation

@software{parcellate,
  author = {Kepler, Gal},
  title = {parcellate: Atlas-based parcellation of neuroimaging data},
  url = {https://github.com/GalKepler/parcellate},
  year = {2024}
}

Acknowledgments

Repository initiated with fpgmaas/cookiecutter-uv.