invertmeeg - A high-level Python library for M/EEG inverse solutions

This package provides 96 inverse solvers for M/EEG source imaging, integrating with the mne-python framework. It covers minimum norm methods, beamformers, Bayesian approaches, sparse recovery, subspace methods, and deep learning models in a unified API.

Read the documentation here.

Installation

The recommended way to install invertmeeg is using uv:

uv add invertmeeg

Alternatively, you can use pip:

pip install invertmeeg

Extras

• Neural Networks: uv add "invertmeeg[ann]" (or pip install "invertmeeg[ann]")
• Visualization: uv add "invertmeeg[viz]" (or pip install "invertmeeg[viz]")

Development

If you are contributing to invertmeeg, we use uv and make to manage the development environment and code quality.

Setup

make install

This will sync all dependencies (including dev tools) and install the pre-commit hooks.

Manual uv setup, install all packages except ann

uv sync --extra viz --extra docs --group dev

Useful Commands

• make lint: Run Ruff and Mypy checks.
• make format: Auto-format code with Ruff.
• make test: Run the test suite with pytest.
• make check: Run all linting and tests.
• make clean: Remove temporary cache files (pycache, .mypy_cache, etc.).

Quick Start

from invert import Solver

# fwd = ...   (mne.Forward object)
# evoked = ... (mne.Evoked object)

solver = Solver("MNE")
solver.make_inverse_operator(fwd)
stc = solver.apply_inverse_operator(evoked)
stc.plot()

Features

• 96 inverse solvers accessible through a single Solver("solver_id") interface
• Interactive benchmark leaderboard -- compare solver performance across datasets, metrics, and categories
• Automatic regularization (GCV, L-curve, product methods)
• Returns standard mne.SourceEstimate objects
• Simulation utilities for benchmarking

Solver Categories

Category	Count	Examples
Minimum Norm	9	MNE, wMNE, dSPM, FISTA, GFT-MNE
LORETA	4	LORETA, sLORETA, eLORETA, SSLOFO
Other Minimum-Norm-like	3	LAURA, Backus-Gilbert, S-MAP
Bayesian	21	Champagne variants, Gamma-MAP, MSP, CMEM, VB-SBL
Beamformers	26	LCMV, DICS, ESMV, SAM, Flex-ESMV, ReciPSIICOS
Dipole Fitting	2	ECD, SESAME
Structured Sparsity	1	Total Variation
Neural Networks	4	FC/ESInet, CovCNN, LSTM, CNN
Matching Pursuit	11	OMP, COSAMP, SOMP, BCS, Subspace Pursuit
MUSIC/Subspace	11	MUSIC, RAP-MUSIC, FLEX-MUSIC, Adaptive-AP
Basis Functions	1	GBF
Other	3	EPIFOCUS, APSE, Random-Noise

Full Algorithm List

Minimum Norm

Full Solver Name	Abbreviation
Minimum Norm Estimate	"mne"
Minimum Norm Estimate with Graph Fourier Transform	"gft-mne"
Weighted Minimum Norm Estimate	"wmne"
Dynamic Statistical Parametric Mapping	"dspm"
Minimum Current Estimate	"l1", "fista", "mce"
Minimum L1 Norm GPT	"gpt", "l1-gpt"
Minimum L1L2 Norm	"l1l2"
GFT Minimum L1 Norm	"gft-l1"
Self-Regularized eLORETA	"sr-eloreta", "self-regularized-eloreta"

LORETA

Full Solver Name	Abbreviation
LORETA	"loreta", "lor"
sLORETA	"sloreta", "slor"
eLORETA	"eloreta", "elor"
SSLOFO	"sslofo"

Other Minimum-Norm-like Algorithms

Full Solver Name	Abbreviation
LAURA	"laura", "laur"
Backus-Gilbert	"backus-gilbert", "b-g", "bg"
S-MAP	"s-map", "smap"

Bayesian

Full Solver Name	Abbreviation
Champagne	"champagne", "champ"
EM Champagne	"em-champagne", "emc"
Convexity Champagne	"convexity-champagne", "coc", "mm-champagne"
MacKay Champagne	"mackay-champagne", "mcc"
TEM Champagne	"tem-champagne", "temc"
AR-EM Champagne	"arem-champagne", "aremc"
Low SNR Champagne	"low-snr-champagne", "lsc"
Adaptive Champagne	"adaptive-champagne", "ac"
Noise Learning Champagne	"nl-champagne", "nlc"
Omni Champagne	"omni-champagne", "oc"
Flex Champagne	"flex-champagne", "fc-champ"
Flex NL Champagne	"flex-nl-champagne", "fnlc"
Gamma-MAP	"gamma-map", "gmap"
Source-MAP	"source-map"
Gamma-MAP-MSP	"gamma-map-msp"
Source-MAP-MSP	"source-map-msp"
Multiple Sparse Priors	"msp"
cMEM	"cmem"
Subspace SBL	"subspace-sbl"
Subspace SBL+	"subspace-sbl-plus"
Variational Bayes SBL	"vb-sbl"

Beamformers

Full Solver Name	Abbreviation
Minimum Variance Adaptive Beamformer	"mvab"
Linearly Constrained Minimum Variance	"lcmv"
Dynamic Imaging of Coherent Sources	"dics"
Standardized Minimum Variance	"smv"
Weight-Normalized Minimum Variance	"wnmv"
Higher-Order Covariance Minimum Variance	"hocmv"
Eigenspace Scalar Minimum Variance	"esmv"
Multiple Constraint Minimum Variance	"mcmv"
Higher-Order Covariance MCMV	"hocmcmv"
Reciprocal PSIICOS (Plain)	"recipsiicos"
Reciprocal PSIICOS (Whitened)	"recipsiicos-whitened"
Synthetic Aperture Magnetometry	"sam"
Empirical Bayesian Beamformer	"ebb"
Adaptive Flexible ESMV	"adapt-flex-esmv"
Flexible ESMV	"flex-esmv"
Deblurring Flexible ESMV	"deblur-flex-esmv"
Safe Flexible ESMV	"safe-flex-esmv"
Sharp Flexible ESMV	"sharp-flex-esmv"
Signal Subspace Projection ESMV	"ssp-esmv"
Iteratively Reweighted ESMV	"ir-esmv"
SSP Iteratively Reweighted ESMV	"ssp-ir-esmv"
Unit Noise Gain	"unit-noise-gain", "ung"

Dipole Fitting

Full Solver Name	Abbreviation
Equivalent Current Dipole	"ecd"
SESAME	"sesame"

Structured Sparsity

Full Solver Name	Abbreviation
Total Variation	"tv", "total-variation", "graph-tv"

Artificial Neural Networks

Full Solver Name	Abbreviation
Fully-Connected Network	"fc", "esinet"
Covariance CNN	"covcnn", "covnet"
Long Short-Term Memory	"lstm"
Convolutional Neural Network	"cnn"

Matching Pursuit / Compressive Sensing

Full Solver Name	Abbreviation
Orthogonal Matching Pursuit	"omp"
Compressive Sampling Matching Pursuit	"cosamp"
Simultaneous Orthogonal Matching Pursuit	"somp"
Random Embedding Matching Pursuit	"rembo"
Subspace Pursuit	"sp"
Structured Subspace Pursuit	"ssp"
Subspace Matching Pursuit	"smp"
Structured Subspace Matching Pursuit	"ssmp"
Subspace-based Subspace Matching Pursuit	"subsmp"
Iterative Subspace-based Subspace Matching Pursuit	"isubsmp"
Bayesian Compressive Sensing	"bcs"

MUSIC/RAP/Subspace

Full Solver Name	Abbreviation
Multiple Signal Classification	"music"
Recursively Applied and Projected MUSIC	"rap-music", "rap"
Truncated Recursively Applied and Projected MUSIC	"trap-music", "trap"
Flexible RAP-MUSIC	"flex-music", "flex"
Flexible RAP-MUSIC 2	"flex-music-2"
Flexible Signal Subspace Matching	"flex-ssm"
Signal Subspace Matching	"ssm"
Flexible Alternating Projections	"flex-ap"
Alternating Projections	"ap"
Adaptive Alternating Projections	"adaptive-ap", "aap"
Generalized Iterative	"generalized-iterative", "gi"

Basis Functions

Full Solver Name	Abbreviation
Geometrically Informed Basis Functions	"gbf"

Other

Full Solver Name	Abbreviation
EPIFOCUS	"epifocus"
APSE	"apse"
Random Noise (Baseline)	"random-noise", "random"

Licensing

This project is dual-licensed:

• Open source: GPLv3.
• Commercial alternative: if you want to use invertmeeg in a proprietary/closed-source product without GPLv3 obligations, a separate commercial license is available (contact below).

For Researchers & Open Source Users (GPLv3)

We love science. This library is free to use for research, education, thesis projects, and open source projects under the terms of GPLv3.

• Attribution required: please credit invertmeeg (and ideally cite this repository) in any publications, reports, or released artifacts that use it.
• Distribution note: if you distribute software that includes/links this library, you need to comply with GPLv3 (including providing corresponding source under GPLv3).

For Industry & Commercial Entities

If you are a company, a startup, or an individual intending to use this library in a proprietary/closed-source product, please contact me to obtain a separate commercial license.

Contact: lukas.hecker.job@gmail.com

Citation

If you use this package and publish results, please cite as:

@Misc{invertmeeg2022,
  author =   {{Lukas Hecker}},
  title =    {{invertmeeg}: A high-level Python library for M/EEG inverse solutions.},
  howpublished = {\url{https://github.com/LukeTheHecker/invertmeeg}},
  year = {since 2022}
}

Send feedback or feature requests to lukas.hecker.job@gmail.com.