Inverse Library

Namespace: INVLIB  ·  Source: src/libraries/inv The Inverse library provides source localization algorithms for MEG/EEG data. All classes reside in the INVLIB namespace.

Algorithm Inventory

Minimum Norm Estimation

Class	Description	MNE-Python	MNE-C
InvMinimumNorm	MNE, dSPM, sLORETA, and eLORETA inverse methods	mne.minimum_norm.apply_inverse	mne_inverse_operator + mne_compute_mne

Supports all four noise-normalization strategies:

Method	Description
MNE	L2-regularized minimum-norm estimate
dSPM	Dynamic Statistical Parametric Mapping — noise-normalized MNE
sLORETA	Standardized Low-Resolution Electromagnetic Tomography
eLORETA	Exact LORETA — iterative weight-normalized MNE for zero localization bias

Beamformers

Class	Description	MNE-Python	MNE-C
InvLCMV	Linearly Constrained Minimum Variance beamformer (time domain)	mne.beamformer.make_lcmv / apply_lcmv	—
InvDICS	Dynamic Imaging of Coherent Sources (frequency domain)	mne.beamformer.make_dics / apply_dics	—
InvBeamformer	Computed spatial filter container	—	—
InvBeamformerCompute	Shared computation routines (data covariance → spatial filter)	—	—

RAP MUSIC

Class	Description	MNE-Python	MNE-C
InvRapMusic	Recursively Applied and Projected Multiple Signal Classification	mne.beamformer.rap_music	—
InvPwlRapMusic	Powell-accelerated RAP MUSIC with gradient-free dipole optimization	—	—
InvDipole	Single dipole result (position, orientation, correlation)	—	—

Contextual Minimum Norm Estimates (CMNE)

Class	Description	MNE-Python	MNE-C
InvCMNE	CMNE inverse: dSPM base estimate refined by LSTM network (ONNX Runtime)	—	—
InvCMNESettings	Configuration for CMNE: ONNX model path, sequence length, LSTM parameters	—	—

Combines classical dSPM with a trained LSTM neural network to incorporate temporal context. At inference time, InvCMNE::compute() runs dSPM to obtain a base source estimate, then applies the ONNX-exported LSTM model sample-by-sample to produce a spatiotemporally refined CMNE result. Falls back to moving-average smoothing if no ONNX model is available.

Reference: Dinh et al. (2021). Contextual Minimum-Norm Estimates (CMNE): A Deep Learning Method for Source Estimation in Neuronal Networks. Frontiers in Neuroscience, 15, 552666. DOI: 10.3389/fnins.2021.552666

Sparse Methods

Class	Description	MNE-Python	MNE-C
InvMxne	Mixed-Norm Estimate (MxNE) — L21-norm sparse inverse via IRLS	mne.inverse_sparse.mixed_norm	—
InvMxneResult	Result container: sparse source estimate, active vertices, iteration count, residual	—	—
InvGammaMap	Gamma-MAP sparse Bayesian learning inverse via EM	mne.inverse_sparse.gamma_map	—
InvGammaMapResult	Result container: sparse source estimate, active vertices, gamma hyperparameters	—	—

MxNE minimizes a group-lasso ($\ell_{2,1}$-norm) cost function using iteratively reweighted least squares with active-set pruning. Gamma-MAP uses an EM algorithm to estimate per-source variance hyperparameters, pruning sources below a configurable threshold.

Dipole Fitting

Class	Description	MNE-Python	MNE-C
InvDipoleFit	High-level driver for sequential ECD fitting	mne.fit_dipole	mne_dipole_fit
InvDipoleFitData	Workspace: sensor geometry, forward model, noise covariance, fit routines	—	libmne internals
InvDipoleFitSettings	Command-line and programmatic settings	—	mne_dipole_fit options
InvDipoleForward	Forward field matrices and SVD for magnetic dipole fitting	—	libfwd
InvGuessData	Precomputed guess-point grid with forward fields	—	libmne
InvEcd	Single equivalent current dipole (position, orientation, amplitude, GOF)	mne.Dipole	—
InvEcdSet	Collection of fitted dipoles	mne.DipoleFixed	—

HPI Fitting

Class	Description	MNE-Python	MNE-C
InvHpiFit	HPI coil dipole fitting for head position tracking	mne.chpi.compute_head_pos	—
InvHpiDataUpdater	Preprocesses raw HPI data (SSP, compensation, model fitting)	mne.chpi.compute_chpi_amplitudes	—
InvHpiModelParameters	Configuration for HPI signal model (line freq, coil freqs, sample rate)	—	—
InvSignalModel	Forward sinusoidal model matrix for HPI coil signals	—	—
InvSensorSet	MEG sensor geometry container for HPI	—	—

Result Types

Class / Enum	Description
InvSourceEstimate	Source estimate container (vertices × time, with time metadata)
tokenize / fromTokens	Tokenization / de-tokenization of an InvSourceEstimate for foundation-model interfacing (inv_source_estimate_token.h)
InvFocalDipole	Focal dipole with free 3D position, moment, and fit-quality metrics
InvSourceCoupling	Correlated grid sources with orientations and correlation matrix
InvConnectivity	Pairwise source connectivity matrix
InvEstimateMethod	Enum: MNE, dSPM, sLORETA, eLORETA, CMNE, LCMV, DICS, SAM, MxNE, GammaMAP, DipoleFit, RapMusic, PwlRapMusic

Doxygen Reference

For method signatures, inheritance diagrams, and source-level documentation see the auto-generated INVLIB namespace in the Doxygen API reference.

See Also

• DSP Library — Signal processing algorithms
• Connectivity Library — Functional connectivity metrics
• MNE Dipole Fit — Application GUI for dipole fitting
• mne_inverse_operator — CLI tool for inverse operator computation