Automated seizure diagnosis system based on feature extraction and channel selection using EEG signals

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Seizures are abnormal brain electrical activity.
• Diagnosis involves methods like EEG, CT, and MRI.
• EEG signals contain numerous features, many irrelevant for diagnosis.

Purpose of the Study:

• To develop an automatic seizure classification system.
• To identify and extract the most significant EEG features for diagnosis.
• To compare the performance of seven different classifiers.

Main Methods:

• Channel selection based on variance to reduce dimensionality.
• Extraction of 11 relevant EEG features per channel.
• Feature averaging, classification, and cross-validation using training/testing sets.
• Testing classifiers using random and continuous case methods.

Main Results:

• KNN classifier showed high precision, specificity, and positive predictability in random case testing.
• Ensemble classifier achieved higher sensitivity and a lower miss-rate (2.3%) in random case testing.
• Ensemble classifier outperformed others in continuous case testing, detecting all seizure cases accurately.

Conclusions:

• The proposed automatic system effectively classifies seizures using significant EEG features.
• The ensemble classifier is highly effective for seizure detection, particularly in continuous monitoring scenarios.
• Feature extraction and selection are crucial for accurate seizure classification from EEG data.