On the interpretation of linear Riemannian tangent space model parameters in M/EEG

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Riemannian tangent space methods achieve high performance in electroencephalography (EEG) and magnetoencephalography (MEG) applications.
• A key limitation is the poor interpretability of these models, hindering biomarker development.
• Existing component-based methods offer better interpretability but may not match performance.

Purpose of the Study:

• To develop a method for transforming linear tangent space model parameters into interpretable patterns.
• To enhance the interpretability of Riemannian tangent space models for EEG/MEG data.
• To facilitate biomarker development by improving model understanding.

Main Methods:

• Proposed a novel method to convert linear tangent space model parameters into interpretable source patterns.
• Utilized typical assumptions to demonstrate the identification of true latent source patterns.
• Validated the approach using simulations and real-world MEG and EEG datasets.

Main Results:

• The proposed method successfully transforms model parameters into interpretable patterns.
• The approach accurately identifies latent source patterns encoding target signals under typical assumptions.
• Riemannian tangent space methods demonstrated robustness to variations in source patterns across observations.

Conclusions:

• The developed method significantly improves the interpretability of Riemannian tangent space models.
• This approach enhances the utility of advanced EEG/MEG methods for biomarker discovery.
• The robustness of Riemannian tangent space methods extends to the derived interpretable patterns.