A Two-Stage Deep Learning Approach for EEG Artifact Removal and Classification: Towards Reliable Wearable...

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) artifact removal is crucial for accurate neural signal processing.
• Ocular artifacts, such as eye blinks and saccadic movements, significantly contaminate EEG data.
• Existing methods often struggle with real-time artifact identification and removal, especially in specific brain regions.

Purpose of the Study:

• To develop and evaluate a novel two-stage system for automated EEG artifact removal and classification.
• To improve the accuracy of artifact removal in temporal and frontal EEG recordings.
• To enable reliable artifact identification for continuous monitoring and Brain-Computer Interface (BCI) applications.

Main Methods:

• A two-stage deep learning approach combining a modified IC-UNet for artifact removal and a modified VGGNet for artifact classification.
• Parallel encoding paths with channel-specific feature extraction in the denoising network.
• Automatic triggering of the classification stage based on signal difference thresholds.

Main Results:

• The denoising network achieved high correlation coefficients between predicted and ground truth signals in temporal (T5: 0.86, T6: 0.85) and frontal (F3: 0.83) regions.
• The classification network demonstrated excellent performance with 99.35% accuracy, correctly classifying 616 out of 620 cases.
• The system effectively identified ocular artifacts, including eye blinks and saccadic movements.

Conclusions:

• The proposed two-stage system offers a feasible and accurate solution for EEG artifact removal and classification.
• This method is particularly relevant for temporal and behind-the-ear EEG recordings, crucial for wearable EEG devices.
• The findings support the development of advanced hybrid BCI systems and continuous EEG monitoring solutions.