Automatic Detection of EEG Epileptiform Abnormalities in Traumatic Brain Injury using Deep Learning

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Biology

Background:

• Traumatic brain injury (TBI) can lead to chronic neurological conditions, including posttraumatic epilepsy (PTE).
• Identifying biomarkers for epileptogenesis, the process leading to seizures after TBI, is crucial for timely intervention.
• Current methods for detecting epileptiform abnormalities (EAs) in TBI patients rely on manual review of electroencephalogram (EEG) data.

Purpose of the Study:

• To develop and validate a deep neural network for automated detection of PTE biomarkers from early TBI EEG data.
• To improve the accuracy and efficiency of identifying patients at risk for developing PTE.
• To lay the groundwork for automated, robust detection of epileptiform activity in TBI patients.

Main Methods:

• Utilized deep neural network architectures, specifically a recurrent neural network (RNN).
• Trained the RNN on continuous EEG data from moderate-to-severe TBI patients within post-injury days 1-7.
• Focused on detecting epileptiform abnormalities (EAs) as potential biomarkers of epileptogenesis.

Main Results:

• The recurrent neural network achieved a highest accuracy of 80.78% in identifying EAs.
• Demonstrated the feasibility of using deep learning for automated analysis of TBI EEG data.
• Highlighted the potential for early detection of PTE risk factors.

Conclusions:

• Automated detection of epileptiform activity in TBI patients using deep neural networks is feasible and accurate.
• This approach can significantly enhance the standard of care by identifying at-risk patients for antiepileptogenic interventions.
• The developed RNN model shows promise for clinical application in TBI monitoring and management.