Incorporating feature selection methods into a machine learning-based neonatal seizure diagnosis.

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Neonatal seizures require accurate and efficient detection methods.
• Electroencephalography (EEG) is crucial for diagnosing neonatal seizures.
• Existing methods may benefit from optimized feature selection for improved classification.

Purpose of the Study:

• To develop a feature selection (FS)-based decision support system for neonatal seizure detection using EEG signals.
• To evaluate the impact of different FS algorithms on classification performance and cost.
• To identify the most effective features and channel differences for accurate seizure classification.

Main Methods:

• Utilized EEG data from 79 term neonates with and without seizures.
• Employed 10 different FS algorithms to reduce feature dimensionality.
• Extracted features from 18 channel differences and assessed classification performance.
• Evaluated classification accuracy using selected feature subsets.

Main Results:

• Feature selection significantly improved classification performance compared to using all extracted features.
• The C4-P4 channel difference achieved the highest classification performance at 98.8%.
• Optimal classification was obtained using only 2 or 3 selected features, demonstrating substantial reduction.

Conclusions:

• Feature reduction using FS algorithms effectively lowers costs and enhances classification performance for neonatal EEG analysis.
• The developed FS-based system shows promise for clinical application in neonatal seizure detection.
• Results provide a foundation for future research in optimizing EEG-based diagnostic tools.