mne_compute_raw_inverse

Overview

mne_compute_raw_inverse computes inverse solutions (MNE, dSPM, or sLORETA) from raw or evoked FIFF data using a pre-computed inverse operator. It supports label-restricted source estimation and outputs results as STC files.

This is a C++ port of the original MNE-C tool by Matti Hämäläinen.

Usage

mne_compute_raw_inverse [options]

Options

Option	Description
--in <file>	Raw or evoked data input file (required)
--inv <file>	Inverse operator file (required)
--snr <value>	SNR value to use for regularization (default: 1.0)
--nave <number>	Number of averages (default: 1 for raw, from data for evoked)
--set <number>	Evoked data set number to process (default: process all)
--bmin <time>	Baseline starting time in milliseconds
--bmax <time>	Baseline ending time in milliseconds
--label <file>	Label file to restrict processing to (can specify multiple)
--labeldir <dir>	Process all labels in directory, compute average waveform per label
--out <file>	Output file name (needed when using --labeldir)
--picknormalcomp	Pick current component normal to cortex only
--spm	Use dSPM noise-normalization method
--sloreta	Use sLORETA noise-normalization method
--mricoord	Output source locations in MRI coordinates
--orignames	Use original label file names in channel names
--align_z	Align waveform signs using surface normal information
--labellist <file>	Output label name list to specified file

Description

This tool applies a pre-computed inverse operator to MEG/EEG data to produce source estimates on the cortical surface. The inverse operator must be pre-computed and stored in a FIFF file.

Inverse Methods

Three methods are available for computing the source estimates:

• MNE (default) — Standard minimum-norm estimate. Provides current amplitude estimates in physical units (Am).

• dSPM (--spm flag) — Dynamic Statistical Parametric Mapping. Produces noise-normalized estimates that are dimensionless statistical test variables. Reduces location bias compared to MNE.

• sLORETA (--sloreta flag) — Standardized Low-Resolution Electromagnetic Tomography. Another noise normalization approach that uses the resolution matrix diagonal for variance estimation.

For mathematical details on these methods, see The Minimum-Norm Estimates.

Regularization (SNR)

The --snr parameter controls the regularization of the inverse solution. The regularization parameter $\lambda^2$ is related to the SNR by $\lambda^2 \approx 1/\text{SNR}^2$.

• Higher SNR → less regularization → noisier but potentially more detailed estimates
• Lower SNR → more regularization → smoother estimates

For averaged evoked data, typical SNR values are 1.0–3.0. The number of averages (--nave) is automatically taken into account.

Label-Based Analysis

The --label option restricts the inverse computation to a specific cortical region defined by a FreeSurfer label file. Multiple labels can be specified. This is useful for ROI-based analyses.

The --labeldir option processes all labels in a directory and computes the average source waveform for each label, which is useful for atlas-based analyses.

Output

The output is written as STC (source estimate) files, which contain the estimated source activity at each source space location over time. These files can be visualized using mne_inspect or other MNE visualization tools.

Examples

Compute dSPM source estimates from evoked data:

mne_compute_raw_inverse \
    --in sample_audvis-ave.fif \
    --inv sample_audvis-meg-oct6-inv.fif \
    --snr 3.0 \
    --spm

Compute MNE estimates restricted to a label:

mne_compute_raw_inverse \
    --in sample_audvis-ave.fif \
    --inv sample_audvis-meg-oct6-inv.fif \
    --snr 2.0 \
    --label auditory-lh.label \
    --picknormalcomp

Process all labels in a directory:

mne_compute_raw_inverse \
    --in sample_audvis-ave.fif \
    --inv sample_audvis-meg-oct6-inv.fif \
    --snr 3.0 --spm \
    --labeldir labels/ \
    --out label_timecourses