Combining sparsity and rotational invariance in EEG/MEG source reconstruction

Area of Science:

• Biomedical Imaging
• Computational Neuroscience
• Signal Processing

Background:

• The electroencephalography (EEG) inverse problem is crucial for non-invasively mapping brain activity.
• Existing methods struggle with accurately localizing multiple, closely spaced neural sources.
• A need exists for robust inverse imaging techniques that are invariant to coordinate systems and promote sparse solutions.

Purpose of the Study:

• Introduce Focal Vector Field Reconstruction (FVR), a novel technique for vector field inverse imaging.
• Address limitations of current methods in EEG source localization.
• Achieve coordinate system invariance and promote sparse solutions for improved accuracy.

Main Methods:

• Developed FVR by defining a regularization penalty as a global l(1)-norm of local l(2)-norms.
• Ensured uniqueness and sparsity of solutions and their spatial derivatives.
• Utilized Earth Mover's Distance (EMD) for comparing source distributions.

Main Results:

• FVR reliably recovered true sources in simulations of 2-3 dipoles, outperforming LORETA and Minimum l(1)-norm (too smooth) and Minimum l(2)-norm (too scattered).
• FVR demonstrated the smallest localization error in both noiseless and noisy simulations based on EMD.
• Applied to real EEG data, FVR accurately localized bilateral somatosensory N20 generators, aligning with neurophysiological knowledge.

Conclusions:

• FVR offers superior performance in EEG inverse problems, particularly for localizing multiple, nearby sources.
• The method's coordinate invariance and sparsity-promoting properties enhance localization accuracy.
• FVR represents a significant advancement for non-invasive brain source imaging.