Improved Manual Annotation of EEG Signals through Convolutional Neural Network Guidance.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Automated artifact handling in electroencephalogram (EEG) signals requires validated algorithms for reliable processing.
• Machine learning advances improve artifact detection but validation is limited by the lack of a gold standard and time-consuming manual annotation.
• Manual annotation of EEG data is prone to errors and is a bottleneck in developing accurate automated artifact detection methods.

Purpose of the Study:

• To propose an iterative learning model to accelerate and reduce errors in manual EEG artifact annotation.
• To enhance the gold standard for EEG artifact annotation by incorporating machine learning feedback.
• To improve the performance of convolutional neural networks (CNNs) for EEG artifact detection.

Main Methods:

• Trained a CNN on expert-annotated resting-state EEG data from typically developing children and children with neurodevelopmental disorders.
• Implemented an iterative approach where the CNN identified poorly learned data for expert re-annotation, refining the gold standard.
• Re-assessed 23% of the data, with experts changing annotations in 25% of cases, and retrained the CNN on the revised dataset.

Main Results:

• The iterative model improved the gold standard by revising incorrectly learned data.
• Training on the expert-revised gold standard increased balanced accuracy from 74% to 80% and precision from 59% to 76%.
• Demonstrated improved separation between artifacts and non-artifacts after retraining the CNN.

Conclusions:

• CNNs show significant promise for enhancing manual annotation of EEG artifacts.
• Iterative learning models can improve the quality of gold-standard datasets for EEG analysis.
• Further improvements in CNN performance are achievable with enhanced gold-standard data.