EEG data alignment across devices using a neural network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate synchronization of multi-device biomedical recordings is crucial for data analysis.
• Existing methods like trigger-based or artifact alignment have practical limitations.

Purpose of the Study:

• To propose and evaluate a novel neural network-based approach for temporal offset estimation between biomedical recordings.
• To achieve synchronization without assuming signal morphological similarities.

Main Methods:

• A neural network model was developed to estimate temporal offsets.
• The model was evaluated on three electroencephalography (EEG) datasets (EESM17, EESM19, Surrey-cEEGrid) with varying hardware and alignment challenges.
• Training was performed on a dataset with perfect synchronization (EESM19).

Main Results:

• The neural network model generalized well to new devices and noisy data.
• The proposed method demonstrated superior performance compared to artifact-based benchmarks.
• Robust temporal alignment was achieved across diverse datasets.

Conclusions:

• The neural network approach provides an effective solution for post hoc synchronization of biomedical data.
• This method overcomes limitations of traditional synchronization techniques.
• It offers a promising tool for analyzing complex, multi-channel biomedical recordings.