Classification of Non-Severe Traumatic Brain Injury from Resting-State EEG Signal Using LSTM Network with ECOC-SVM.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Traumatic brain injury (TBI) is a common injury from head impacts, frequently leading to insurance claims.
• Insurance fraud involving false TBI claims necessitates reliable and rapid diagnostic systems.
• Current imaging methods like CT and MRI have limitations, including immobility issues.

Purpose of the Study:

• To develop an instant brain condition classification system for non-severe TBI patients and healthy subjects.
• To utilize resting-state electroencephalogram (EEG) as a non-invasive input for classification.
• To overcome the immobility constraints associated with CT and MRI scans.

Main Methods:

• A novel classification architecture combining Long Short-Term Memory (LSTM) and Error-Correcting Output Coding Support Vector Machine (ECOC-SVM).
• Inputting pre-processed EEG time series data to the LSTM network, which remembers information from previous time steps.
• Training the ECOC-SVM using activations from the LSTM cell for multiclass classification.

Main Results:

• The proposed architecture achieved a 100% classification accuracy.
• The system effectively amplified the temporal advantages of EEG signals.
• Performance metrics including classification accuracy, sensitivity, specificity, and precision surpassed existing methods.

Conclusions:

• The developed LSTM-ECOC-SVM architecture provides a highly accurate and efficient method for classifying non-severe TBI.
• Resting-state EEG offers a viable, non-invasive alternative to traditional neuroimaging for TBI assessment.
• This system addresses the need for instant brain condition classification, potentially mitigating insurance fraud.