Deep Learning-Based Localization of EEG Electrodes Within MRI Acquisitions.

Area of Science:

• Neuroimaging
• Medical Physics
• Computational Neuroscience

Background:

• Simultaneous electroencephalographic (EEG) and functional magnetic resonance imaging (fMRI) offers high spatial and temporal resolution for brain activity measurement.
• This bimodal neuroimaging is particularly valuable for epilepsy research, aiding in the localization of epileptic networks.
• Accurate EEG electrode positioning is crucial for resolving the complex inverse problem in EEG source localization, yet current methods often rely on imprecise fiducial points.

Purpose of the Study:

• To develop and validate a fully automatic method using neural networks for precise EEG electrode localization from ultra-short echo-time MRI.
• To improve the accuracy of EEG source localization in simultaneous EEG-fMRI studies.
• To enhance the analysis of brain activity, especially in epilepsy research.

Main Methods:

• A novel, fully automatic method employing neural networks for segmenting ultra-short echo-time MR volumes.
• The method involves two key steps: neural network-based image segmentation and subsequent registration of an EEG template to detected electrode positions.
• The neural network was trained on 37 MR volumes and tested on 23 new volumes.

Main Results:

• The automated method achieved an average EEG electrode detection accuracy of 99.7%.
• The average positional error for detected electrodes was 2.24 mm.
• The method demonstrated 100% accuracy in labeling the detected EEG electrodes.

Conclusions:

• The proposed neural network-based method provides a highly accurate and automatic solution for EEG electrode localization in simultaneous EEG-fMRI.
• This advancement can significantly improve the reliability of EEG source localization and epilepsy network analysis.
• The technique offers a robust tool for multimodal neuroimaging, enhancing the understanding of brain function and dysfunction.