BandRouteNet: Adaptive Neural Network for EEG Artifact Removal
Researchers have introduced BandRouteNet, a novel adaptive frequency-aware neural network designed to remove artifacts from Electroencephalography (EEG) signals. EEG data is frequently contaminated by interference such as electrooculographic (EOG) and electromyographic (EMG) noise, which compromises signal quality for neurological diagnosis and brain-computer interfaces. BandRouteNet addresses this by combining band-specific processing with full-band contextual modeling. It employs a routing mechanism to adaptively determine denoising intensity across temporal locations within specific frequency bands, while a parallel full-band conditioner extracts global context to refine the signal. Extensive experiments on the EEGDenoiseNet benchmark dataset demonstrate that BandRouteNet outperforms existing methods in terms of Relative Root Mean Square Error and Signal-to-Noise Ratio Improvement under various artifact conditions. Notably, the model achieves high performance with only 0.2 million trainable parameters, making it highly parameter-efficient. This efficiency highlights its potential for deployment in resource-constrained applications, offering a robust solution for enhancing EEG signal reliability in clinical and technical settings.
Wire timeline
BandRouteNet: Adaptive Neural Network for EEG Artifact Removal
Researchers have introduced BandRouteNet, a novel adaptive frequency-aware neural network designed to remove artifacts from Electroencephalography (EEG) signals. EEG data is frequently contaminated by interference such as electrooculographic (EOG) and electromyographic (EMG) noise, which compromises signal quality for neurological diagnosis and brain-computer interfaces. BandRouteNet addresses this by combining band-specific processing with full-band contextual modeling. It employs a routing mechanism to adaptively determine denoising intensity across temporal locations within specific frequency bands, while a parallel full-band conditioner extracts global context to refine the signal. Extensive experiments on the EEGDenoiseNet benchmark dataset demonstrate that BandRouteNet outperforms existing methods in terms of Relative Root Mean Square Error and Signal-to-Noise Ratio Improvement under various artifact conditions. Notably, the model achieves high performance with only 0.2 million trainable parameters, making it highly parameter-efficient. This efficiency highlights its potential for deployment in resource-constrained applications, offering a robust solution for enhancing EEG signal reliability in clinical and technical settings.
cs.AI updates on arXiv.org