Single-channel electroencephalography decomposition by detector-atom network and its pre-trained model.

Area of Science:

• Neuroscience
• Signal Processing
• Artificial Intelligence

Background:

• Multi-channel spatial features are crucial for analyzing electroencephalography (EEG) signals.
• Existing methods often require extensive channel data for signal decomposition.
• Limited channel availability poses a challenge in practical EEG analysis.

Purpose of the Study:

• To develop a novel single-channel decomposition approach for EEG signals.
• To overcome limitations of multi-channel feature dependency in signal analysis.
• To enable effective EEG signal decomposition with minimal channel data.

Main Methods:

• Proposed a model based on the hypothesis that EEG signals consist of short, shift-invariant waves (atoms).
• Designed an artificial neural network (ANN) as a decomposer to estimate these atoms.
• The ANN detects time shifts and amplitude modulations of the identified atoms within the signal.

Main Results:

• Validated the method's efficacy across diverse brain-computer interface and neuroscience scenarios.
• Demonstrated enhanced performance in signal analysis and denoising.
• Cross-dataset validation confirmed the feasibility of a pre-trained, plug-and-play module.

Conclusions:

• The novel single-channel decomposition method effectively analyzes EEG signals without multi-channel features.
• The ANN-based decomposer accurately estimates signal atoms, their shifts, and modulations.
• The approach offers a versatile and efficient solution for EEG analysis, applicable across datasets.