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

Le résumé fourni par la source

Overview CaliBrain v1.0.0 marks a major consolidation of the package around a stable end-to-end workflow for uncertainty estimation and calibration in EEG/MEG inverse source imaging. This release promotes the current development pipeline to the supported baseline and formalizes the workflow for data generation, aggregation, calibration, and calibration-figure generation. Highlights Consolidated the supported inverse-solver stack around gamma_map_sflex, gamma_lambda_map_sflex, BMN, and BMN_joint Replaced the legacy benchmarking workflow with DataGenerator Added explicit workflow modules for: data generation aggregation calibration Reworked calibration from within-subject, across-source fitting to pooled source datasets across subjects Changed isotonic regression fitting to operate on subject-level splits rather than one subject at a time Added support for uncertainty modes pointwise and aggregated Added support for free-orientation interval types full_cov and marginal Added support for calibration modes precal, post_oracle, post_pooled, post_pooled_mismatch, and post_fixed Expanded metric evaluation support for mse, mae, rmse, rmae, mean_posterior_std, emd, mean_signed_deviation, mean_absolute_deviation, max_underconfidence_deviation, and max_overconfidence_deviation Standardized manifest-based discovery and downstream aggregated dataset writing for calibration workflows Updated the documentation and changelog to reflect the supported workflow and version baseline Breaking changes Removed the legacy benchmarking-based workflow in favor of DataGenerator Changed calibration fitting from single-subject source-wise calibration to pooled cross-subject calibration with subject-split isotonic regression Standardized the active pipeline around manifest-driven aggregation and calibration outputs Removed or deprecated older unsupported solver and method branches from the supported workflow Workflow summary calibrain/workflows/data_generation.py Runs end-to-end synthetic experiment generation from configuration files Resolves solver grids and writes posterior summaries plus manifest entries calibrain/workflows/aggregation.py Reads posterior summaries from the manifest Filters runs and writes compact calibration-ready datasets calibrain/workflows/calibration.py Loads aggregated train/eval datasets Fits and evaluates calibration mappings and writes calibration summaries calibrain/workflows/plot_paper_calibration_figures.py Collects calibration outputs Builds paper-style fixed/free calibration comparison figures Supported calibration modes precal post_oracle post_pooled post_pooled_mismatch post_fixed Supported uncertainty settings Modes pointwise aggregated Free-orientation interval types full_cov marginal Notes This release establishes the 1.0.0 workflow baseline on main Documentation and CI now target Python >=3.10

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.