A lightweight deep convolutional neural network for detecting artifacts in continuous EEG signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for neurological diagnostics.
• Artifacts like eye movements, muscle activity, and non-physiological signals can compromise EEG data quality.
• Automated artifact detection is essential for reliable EEG analysis.

Purpose of the Study:

• To develop and validate lightweight, artifact-specific convolutional neural networks (CNNs) for automated detection of EEG artifacts.
• To determine optimal temporal windows for classifying different artifact types (eye movement, muscle, non-physiological).

Main Methods:

• Trained three binary CNN detectors on the Temple University Hospital EEG artifact corpus using patient-level data splits.
• Standardized EEG signals to 250 Hz with a 22-channel bipolar montage.
• Evaluated non-overlapping signal segments ranging from 1 to 30 seconds, optimizing operating points using Youden's J statistic.

Main Results:

• CNNs demonstrated superior performance compared to rule-based detection methods.
• Optimal detection windows varied by artifact class: 20s for eye movements (ROC AUC 0.975), 5s for muscle artifacts (accuracy 93.2%), and 1s for non-physiological artifacts (F1 0.774).

Conclusions:

• Artifact-specific CNNs with class-tailored temporal windows offer a reliable approach to EEG artifact detection.
• This method surpasses traditional rule-based performance, providing practical guidance for clinical EEG quality control.