CaliBrain Documentation#

CaliBrain is a Python framework for uncertainty estimation and calibration in EEG/MEG inverse source imaging.

commits Documentation (latest)

Overview#

CaliBrain supports both:

  • Regression (continuous source estimates)

  • Classification (binary activation detection)

Key Features:

  • Setup of source space, BEM model, forward solution, and leadfield matrices

  • Simulation of source activity and sensor-level measurements with controllable noise and source orientation (fixed or free)

  • Solving the inverse problem and reconstructing source time courses

  • Estimation and visualization of confidence intervals

  • Calibration analysis by comparing expected vs. observed confidence levels

Supported Inverse Methods#

  • Gamma-MAP

  • eLORETA

  • Bayesian Minimum Norm

Calibration Tasks#

1. Regression (Confidence Interval Calibration)#

  • Check if true simulated source currents fall within predicted confidence intervals

  • Plot calibration curve (Expected vs. Observed coverage)

  • Well-calibrated models should follow the diagonal

2. Classification (Activation Calibration)#

  • Assess if estimated activation probabilities match true activation frequencies

  • Plot calibration curve for activation detection

  • Ideal calibration follows the diagonal

Main Parameters#

  • Estimator: Gamma-MAP, eLORETA, Bayesian Minimum Norm

  • Orientation: Fixed or Free

  • Noise Type: Oracle, Baseline, Cross-Validation, Joint Learning

  • SNR Level (α): Control regularization strength

  • Active Sources (nnz): Number of nonzero sources

Uncertainty Calibration Parameters

Outcomes#

  • Regression Calibration Curves (confidence intervals)

  • Classification Calibration Curves (activation probabilities)

  • Quantitative Calibration Metrics

Package Components#

CaliBrain is built around a modular architecture:

  • LeadfieldBuilder - Creates forward models

  • SourceSimulator - Generates brain activity

  • SensorSimulator - Simulates measurements

  • SourceEstimator - Solves inverse problems

  • UncertaintyEstimator - Quantifies uncertainty

  • MetricEvaluator - Evaluates performance

  • Visualizer - Creates visualizations

  • Benchmark - Orchestrates workflows

License and Citation#

This project is licensed under the GNU Affero General Public License v3.0. See LICENSE.

If you use CaliBrain in your research, please cite relevant works in EEG/MEG source imaging and uncertainty quantification.

Indices and Tables#