Interictal Epileptiform Discharge Detection Using Time-Frequency Analysis and Transfer Learning.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Interictal epileptiform discharges (IEDs) are key indicators in epilepsy diagnosis.
• Manual analysis of long electroencephalogram (EEG) signals is time-consuming and prone to error.
• Automated IED detection can assist clinicians by identifying cortical irritations and predicting seizures.

Purpose of the Study:

• To develop and evaluate a transfer-learning-based deep learning model for automated IED detection.
• To analyze time-frequency representations of IEDs from scalp EEG data.
• To improve the efficiency and accuracy of epilepsy diagnosis.

Main Methods:

• Utilized a deep residual network (ResNet) fine-tuned with transfer learning.
• Analyzed time-frequency representations of scalp EEG data.
• Evaluated the model on the Temple University Events EEG dataset for binary classification of IEDs.

Main Results:

• Achieved a promising F1-score of 88.52% for binary classification of IEDs.
• Demonstrated the effectiveness of the transfer-learning approach in analyzing EEG data.
• The model shows potential for assisting in clinical epilepsy diagnosis.

Conclusions:

• The proposed transfer-learning deep residual network offers an effective method for automated IED detection.
• This approach can significantly reduce the burden of manual EEG analysis for clinicians.
• The findings support the use of advanced machine learning techniques in epilepsy management.