Unsupervised Event Characterization and Detection in Multichannel Signals: An EEG application

Area of Science:

• Biomedical Engineering
• Signal Processing
• Neuroscience

Background:

• Artifacts in electroencephalogram (EEG) recordings can obscure neural activity.
• Accurate event detection is crucial for analyzing brain signals.
• Existing methods often require supervised learning and predefined event knowledge.

Purpose of the Study:

• To develop a novel unsupervised method for automatic event characterization and detection in multichannel signals.
• To apply this method for identifying artifacts in EEG recordings.
• To evaluate the performance of the unsupervised method against a supervised approach.

Main Methods:

• An unsupervised algorithm was developed to automatically characterize and detect events in multichannel signals.
• The method identifies artifacts in electroencephalogram (EEG) data.
• Performance was evaluated by comparing it to a supervised method, with a specific example demonstrating artifact detection.

Main Results:

• The unsupervised method achieved classification performance comparable to supervised methods.
• It successfully detected events without requiring training data.
• The algorithm can identify unknown events in signals and provides an optimal detection window.

Conclusions:

• The proposed unsupervised method offers a robust alternative for event detection in multichannel signals, particularly EEG.
• It overcomes limitations of supervised methods by not needing training data or prior knowledge of events.
• The real-time applicability and optimal windowing enhance its utility in various signal processing applications.