Machine learning for detection of interictal epileptiform discharges

Area of Science:

• Neurology
• Biomedical Engineering
• Medical Informatics

Background:

• Electroencephalogram (EEG) is crucial for epilepsy diagnosis and classification.
• Interictal Epileptiform Discharges (IEDs) indicate seizure likelihood but visual analysis is time-consuming and subjective.
• High misdiagnosis rates necessitate automated IED detection methods.

Purpose of the Study:

• To review and discuss various automated approaches for Interictal Epileptiform Discharge (IED) detection in EEG.
• To analyze the performance and limitations of different IED detection methodologies.
• To highlight the need for standardization in datasets and outcome measures for clinical implementation.

Main Methods:

• Review of automated IED detection techniques, including mimetic methods, traditional machine learning, and deep learning.
• Discussion of the evolution of automated IED detection research over the past 45 years.
• Comparative analysis of different approaches based on reported performance and limitations.

Main Results:

• Traditional machine learning and deep learning methods have demonstrated the most promising results for automated IED detection.
• The application of these advanced methods in clinical practice is expanding.
• Significant variability exists in current methodologies, hindering direct comparison.

Conclusions:

• Automated IED detection in EEG shows significant potential to improve epilepsy diagnosis accuracy and efficiency.
• Further research and standardization are required to validate and implement these methods in clinical settings.
• Objective comparison of algorithms is essential for selecting reliable automated tools for epilepsy management.