EEG-AI: An agentic system for AI-assisted semi-automated EEG preprocessing and artifact removal.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Electroencephalography (EEG) is crucial for identifying neural markers and personalizing treatments.
• Traditional EEG preprocessing methods like Independent Component Analysis (ICA) are time-consuming due to manual artifact inspection.
• Low signal-to-noise ratio and artifacts often complicate EEG data interpretation.

Purpose of the Study:

• To develop an AI-driven, human-in-the-loop framework for EEG preprocessing.
• To enhance the efficiency and accuracy of artifact removal in EEG signals.
• To integrate AI agents with expert oversight for iterative refinement of artifact correction.

Main Methods:

• A large language model (LLM)-driven agent was developed to manage complex EEG signal mixtures.
• The framework employs an iterative reasoning loop with a closed-loop policy for adaptive artifact correction.
• The AI agent interprets probabilistic outputs from classifiers, guiding component selection and re-analysis.

Main Results:

• The AI agent system outperformed conventional preprocessing pipelines in signal cleaning.
• Achieved high accuracy in artifact detection and ICA classification compared to expert baselines (Person's r = 0.666 ± 0.188).
• Demonstrated improved reconstruction quality with minimal error (RMSE = 5×10⁻⁶ ± 1×10⁻⁶).

Conclusions:

• The AI agent framework streamlines EEG preprocessing, reducing manual effort.
• Maintains expert oversight through a closed-loop policy for reliable artifact removal.
• Ensures reproducibility and auditability in EEG data analysis.