CaliBrain#

PyPI version Documentation status Supported Python versions License DOI

CaliBrain: A Python toolbox for uncertainty estimation and calibration in EEG/MEG inverse source imaging.

Overview#

Inverse source imaging is an ill-posed problem: different source configurations can explain the same sensor data. CaliBrain addresses a central question in Bayesian source imaging: are posterior uncertainty estimates empirically reliable? The toolbox provides simulation-based workflows for generating source activity, propagating it through forward models, reconstructing posterior source estimates, quantifying empirical coverage, and evaluating recalibration maps under controlled experimental conditions.

CaliBrain pipeline overview

Documentation#

See the tutorials, API reference, and workflow guides in the navigation menu for runnable examples and detailed usage.

Citation#

If you use CaliBrain in academic work, please cite the software archive:

Orabe, Mohammad, Huseynov, Ismail T., Nagarajan, Srikantan, & Haufe, Stefan. (2026). CaliBrain: A Python toolbox for uncertainty estimation and calibration in EEG/MEG inverse source imaging (v1.0.2). Zenodo. https://doi.org/10.5281/zenodo.20721580

Workflow#

The package follows this workflow:

  1. generate source-level ground truth under controlled sparsity and amplitude assumptions;

  2. project sources to sensors through a leadfield and add noise at defined SNR;

  3. reconstruct posterior means and uncertainty summaries with inverse solvers;

  4. convert uncertainty summaries into intervals, ellipses, or ellipsoids;

  5. compare empirical against nominal coverage;

  6. fit isotonic recalibration functions on training splits and evaluate them on held-out splits.

CaliBrain currently supports fixed and free-orientation source models for inverse source imaging methods:

  • gamma_map_sflex for Gamma-MAP reconstruction with sparse basis field expansions;

  • gamma_lambda_map_sflex for the S-FLEX Gamma-MAP variant with joint sparsity and lambda regularization;

  • BMN as a Bayesian minimum norm baseline;

  • BMN_joint as a Bayesian minimum norm variant with joint gamma/lambda learning.

Installation#

From PyPI:

python -m pip install calibrain

From a local checkout:

git clone https://github.com/braindatalab/CaliBrain.git
cd CaliBrain
python -m pip install -e .

License#

CaliBrain is distributed under the BSD 3-Clause License. See LICENSE.