Bidirectional Mamba-2 boosts EEG super-resolution via regression and diffusion

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Electroencephalography (EEG) offers high temporal resolution for brain activity but suffers from limited spatial detail due to sparse electrode placement.
• Improving spatial resolution in EEG is crucial for detailed neuroimaging and clinical diagnostics.

Purpose of the Study:

• To develop a novel framework for spatio-temporal super-resolution of EEG signals using low-density recordings.
• To enhance the anatomical coherence and contextual awareness of reconstructed high-resolution EEG data.

Main Methods:

• Introduced DiBiMa-EEGSR, a bidirectional Mamba-2 diffusion framework for EEG super-resolution.
• Formulated super-resolution as conditional generative inference, integrating diffusion with a bidirectional state-space model.
• Utilized low-resolution inputs, electrode positions, and task labels for conditioning, and employed a one-step sampling strategy for efficient inference.

Main Results:

• DiBiMa-EEGSR significantly improved reconstruction accuracy, spatial coherence, and spectral preservation compared to existing methods.
• The framework demonstrated superior performance in both spatial and temporal upsampling tasks across public benchmarks.
• Achieved high-resolution EEG reconstruction without requiring additional hardware.

Conclusions:

• DiBiMa-EEGSR provides a scalable and effective solution for high-resolution electrophysiological imaging from standard EEG.
• The proposed method advances EEG analysis by enabling more detailed insights into brain function and dysfunction.
• The framework offers a promising pathway for improved non-invasive neuroimaging techniques.