Deep Learning Recurrent Neural Network for Concussion Classification in Adolescents Using Raw Electroencephalography...

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Artificial neural networks (ANNs) show promise in medical decision support, particularly in neuroscience and neuroimaging.
• Classifying concussion using electroencephalography (EEG) data with ANNs is an active research area.
• Current methods often require specialized, research-grade equipment, limiting clinical application.

Purpose of the Study:

• To develop a clinically practical concussion classification system using a minimal subset of EEG sensors.
• To investigate the feasibility of using fewer EEG channels for accurate concussion detection.
• To reduce the reliance on extensive and costly neuroimaging equipment.

Main Methods:

• Utilized a deep learning long short-term memory (LSTM) recurrent neural network architecture.
• Analyzed resting-state EEG data from 23 athletes with concussion and 35 control athletes.
• Ranked 64 EEG channels by contribution to classification and selected a minimal subset.

Main Results:

• A classifier trained on the top six EEG channels achieved 94% accuracy in identifying concussions.
• Demonstrated high accuracy comparable to systems using a larger number of sensors.
• Identified specific EEG channels crucial for effective concussion classification.

Conclusions:

• Concussion classification is feasible using raw, resting-state EEG data from a small number of sensors.
• This research is a significant step toward developing portable and user-friendly EEG systems for clinical concussion assessment.
• Minimal EEG sensor systems can provide accurate and accessible concussion detection.