Epilepsy-IEDs: An automated machine learning model for detecting interictal epileptiform discharges from scalp...

Area of Science:

• Neurology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Visual electroencephalogram (EEG) analysis for diagnosing epilepsy is subjective and time-consuming.
• Interictal epileptiform discharges (IEDs) are critical diagnostic markers in epilepsy.
• Automating IED detection can enhance diagnostic accuracy and efficiency.

Purpose of the Study:

• To develop and validate an automated machine learning model for detecting IEDs in EEG.
• To compare the performance of different algorithms for IED detection.
• To create an interpretable and clinically applicable tool for epilepsy diagnosis.

Main Methods:

• Developed an automated IED detection model (Epilepsy-IEDs) using machine learning.
• Trained the model on 141 scalp EEG recordings with 2,597 IEDs and 4,633 non-IEDs.
• Evaluated model performance using four algorithms, including Extreme Gradient Boosting (XGBoost), with separate daytime analysis.
• Assessed generalization on held-out epilepsy patients and specificity on a non-epilepsy cohort.

Main Results:

• The XGBoost-based Epilepsy-IEDs model achieved high sensitivity (84.6%-87.1%) and AUC (0.966-0.973).
• The model demonstrated robust generalization in epilepsy patients (AUC = 0.878-0.890) and good specificity (71.54%) in non-epilepsy subjects.
• A simplified 10-feature variant maintained strong performance (AUC = 0.959).

Conclusions:

• The Epilepsy-IEDs model offers an accurate and interpretable solution for automated IED detection.
• The streamlined version is suitable for integration into clinical EEG workflows.
• Automated IED detection can significantly aid in epilepsy diagnosis.