INVLIB::InvCMNE Class Reference

CMNE inverse solver. More...

#include <inv_cmne.h>

Static Public Member Functions
static InvCMNEResult	compute (const Eigen::MatrixXd &matEvoked, const Eigen::MatrixXd &matGain, const Eigen::MatrixXd &matNoiseCov, const Eigen::MatrixXd &matSrcCov, const InvCMNESettings &settings)
static bool	applyCmne (const Eigen::MatrixXd &matDspmData, const QString &onnxModelPath, Eigen::MatrixXd &sensing, Eigen::MatrixXd &prediction, Eigen::MatrixXd &cmne)
static Eigen::MatrixXd	controlEstimate (const Eigen::MatrixXd &matDspmData, int lookBack)
static Eigen::MatrixXd	zScoreRectify (const Eigen::MatrixXd &matStcData)
static UTILSLIB::PythonRunnerResult	trainLstm (const QString &fwdPath, const QString &covPath, const QString &epochsPath, const QString &outOnnxPath, const InvCMNESettings &settings, const QString &gtStcPrefix={}, int hiddenSize=256, int numLayers=1, int trainEpochs=50, double learningRate=1e-3, int batchSize=64, const QString &finetuneOnnxPath={}, const QString &pythonExe=QStringLiteral("python3"))

Detailed Description

CMNE inverse solver.

Contextual Minimum Norm Estimate (CMNE) inverse solver.

Implements the algorithm from: Dinh et al. "Contextual Minimum-Norm Estimates (CMNE): A Deep Learning Method for Source Estimation in Neuroimaging", 2021.

Definition at line 92 of file inv_cmne.h.

Member Function Documentation

applyCmne()

bool InvCMNE::applyCmne ( const Eigen::MatrixXd & matDspmData, const QString & onnxModelPath, Eigen::MatrixXd & sensing, Eigen::MatrixXd & prediction, Eigen::MatrixXd & cmne )

static

Contextual estimate of Eqs. 9-13, as cmne.apply_cmne with normalised weights.

The model must be exported by cmne.export_onnx: look-back, source count and rectification are read from its cmne_config metadata.

Parameters

[in]	matDspmData	Signed dSPM estimate (n_sources x n_times).
[in]	onnxModelPath	Path to the ONNX model.
[out]	sensing	q_t (Eq. 9).
[out]	prediction	LSTM prediction (q_t for the first k samples).
[out]	cmne	Contextual estimate b_t.

Returns

False if the model cannot be loaded, does not fit the data, or there are no more than k samples.

Definition at line 197 of file inv_cmne.cpp.

compute()

InvCMNEResult InvCMNE::compute ( const Eigen::MatrixXd & matEvoked, const Eigen::MatrixXd & matGain, const Eigen::MatrixXd & matNoiseCov, const Eigen::MatrixXd & matSrcCov, const InvCMNESettings & settings )

static

Compute CMNE inverse solution.

Parameters

[in]	matEvoked	Evoked data (n_channels x n_times).
[in]	matGain	Forward gain matrix (n_channels x n_sources).
[in]	matNoiseCov	Noise covariance (n_channels x n_channels).
[in]	matSrcCov	Source covariance (n_sources x n_sources, diagonal).
[in]	settings	CMNE settings.

Returns

InvCMNEResult containing dSPM and CMNE source estimates.

Definition at line 68 of file inv_cmne.cpp.

controlEstimate()

MatrixXd InvCMNE::controlEstimate ( const Eigen::MatrixXd & matDspmData, int lookBack )

static

Control estimate of the paper (cmne.control_estimate): q_t times the mean of q over the previous lookBack samples, without an LSTM.

Parameters

[in]	matDspmData	Signed dSPM estimate (n_sources x n_times).
[in]	lookBack	Window length k.

Returns

Control estimate (n_sources x n_times).

Definition at line 185 of file inv_cmne.cpp.

trainLstm()

UTILSLIB::PythonRunnerResult InvCMNE::trainLstm ( const QString & fwdPath, const QString & covPath, const QString & epochsPath, const QString & outOnnxPath, const InvCMNESettings & settings, const QString & gtStcPrefix = {}, int hiddenSize = 256, int numLayers = 1, int trainEpochs = 50, double learningRate = 1e-3, int batchSize = 64, const QString & finetuneOnnxPath = {}, const QString & pythonExe = QStringLiteral("python3") )

static

Train the CMNE LSTM model by invoking the Python training script.

This is a convenience wrapper that calls scripts/ml/training/train_cmne_lstm.py via UTILSLIB::PythonRunner. The heavy lifting (PyTorch LSTM training + ONNX export) happens in Python; C++ only launches the process and streams its output.

Parameters

[in]	fwdPath	Path to forward solution FIFF file.
[in]	covPath	Path to noise covariance FIFF file.
[in]	epochsPath	Path to epochs FIFF file.
[in]	outOnnxPath	Desired output path for the ONNX model.
[in]	settings	CMNE settings (look-back, method, SNR are forwarded).
[in]	gtStcPrefix	Ground-truth STC prefix (optional; empty = simulation mode).
[in]	hiddenSize	LSTM hidden dimension (default 256).
[in]	numLayers	LSTM layers (default 1).
[in]	trainEpochs	Number of training epochs (default 50).
[in]	learningRate	Learning rate (default 1e-3).
[in]	batchSize	Batch size (default 64).
[in]	finetuneOnnxPath	Existing ONNX model to fine-tune from (optional).
[in]	pythonExe	Python interpreter (default "python3").

Returns

PythonRunnerResult with exit code, captured output and progress.

Definition at line 245 of file inv_cmne.cpp.

zScoreRectify()

MatrixXd InvCMNE::zScoreRectify ( const Eigen::MatrixXd & matStcData )

static

Rectify and z-score each source over time (Eq. 9).

Parameters

[in]

matStcData

Source data (n_sources x n_times).

Returns

|x| standardised per row; constant rows are centred only.

Definition at line 178 of file inv_cmne.cpp.

---

The documentation for this class was generated from the following files:

• src/libraries/inv/minimum_norm/inv_cmne.h
• src/libraries/inv/minimum_norm/inv_cmne.cpp