Related Experiment Video
Updated: Jun 12, 2025
11:28
Concurrent EEG and Functional MRI Recording and Integration Analysis for Dynamic Cortical Activity Imaging
Published on: June 30, 2018
11.6K
Localized estimation of event-related neural source activity from simultaneous MEG-EEG with a recurrent neural
Jamie A O'Reilly1, Judy D Zhu2, Paul F Sowman3
1School of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangkok, 10520, Thailand.
Summary
This study introduces a recurrent neural network (RNN) for estimating brain activity from MEEG data, outperforming traditional methods by capturing temporal dynamics. The RNN offers a novel, data-driven approach to understanding neural signal generation.
Area of Science:
- Neuroimaging
- Computational Neuroscience
- Biophysics
Background:
- Estimating intracranial current sources from extracranial MEEG signals is a complex inverse problem.
- Existing methods often neglect the temporal dynamics of event-related brain processes by analyzing time samples independently.
Purpose of the Study:
- To develop and evaluate a recurrent neural network (RNN) for intracranial current source estimation using simultaneously recorded MEEG data.
- To compare the RNN's performance against established methods like eLORETA and MxNE.
- To investigate the impact of different training labels and source space configurations on estimation accuracy.
Main Methods:
- A recurrent neural network (RNN) was developed for MEEG source estimation, incorporating sequential data relationships.
- The RNN underwent two training phases: pre-training and transfer learning with L1 regularization.
- Performance was evaluated using MEEG, MEG, and EEG labels, comparing volumetric and surface source spaces, and contrasting with eLORETA and MxNE.
Main Results:
- The RNN approach significantly outperformed eLORETA and MxNE across signal-to-noise ratio, correlation, and mean-squared error metrics.
- Using MEEG labels with fixed-orientation surface sources yielded the most consistent and accurate source estimates.
- The RNN successfully generated temporal dynamics, modeling the transformation from events to neural signals.
Conclusions:
- The RNN provides a powerful, data-driven method for MEEG source reconstruction, effectively capturing temporal dynamics crucial for understanding brain processes.
- This approach offers a unique advantage over traditional methods by modeling the sequential nature of neural data.
- The findings highlight the potential of deep learning for advancing non-invasive neuroimaging analysis.
