Robin's Viewer: Using deep-learning predictions to assist EEG annotation.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Automated electroencephalogram (EEG) annotation using machine learning (ML) aids in tasks like artifact recognition and seizure detection.
• Manual EEG annotation is susceptible to bias and lacks efficiency.
• Existing automated methods prevent user inspection and correction of model predictions.

Purpose of the Study:

• To develop an interactive EEG viewer that integrates deep learning model predictions with human expertise for improved annotation.
• To address the limitations of purely automated or manual EEG annotation processes.

Main Methods:

• Developed Robin's Viewer (RV), a Python-based, open-source, interactive web application for time-series EEG data annotation.
• Integrated deep learning model prediction visualization into the EEG viewer.
• Built upon Plotly, Dash, and the MNE-Python toolbox, supporting common EEG file formats.

Main Results:

• RV visualizes deep learning model predictions alongside EEG data, allowing users to review and re-evaluate annotations.
• The viewer incorporates standard EEG annotation features, including artifact marking and preprocessing customization.
• Facilitates a hybrid approach combining AI predictions with expert knowledge.

Conclusions:

• Robin's Viewer enhances EEG annotation by merging the predictive capabilities of deep learning with the critical judgment of scientists and clinicians.
• This tool optimizes the EEG annotation workflow, with potential for future expansion to detect clinical patterns like sleep stages and abnormalities.