Automatic epilepsy detection using fractal dimensions segmentation and GP-SVM classification.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Epilepsy seizure detection relies heavily on accurate feature extraction from electroencephalographic (EEG) data.
• Traditional methods face challenges in maximizing class separability and reducing data dimensionality.

Purpose of the Study:

• To develop novel feature extraction and automatic epilepsy seizure classification methods.
• To combine machine learning with genetic evolution algorithms for improved EEG analysis.

Main Methods:

• Preprocessing EEG signals using digital filtration and fractal dimension-based adaptive segmentation.
• Employing genetic programming (GP) with support vector machine (SVM) confusion matrix as a fitness function for feature extraction and classification.

Main Results:

• The GP-SVM method significantly improves classification performance by reducing feature dimensionality.
• Automatic determination of feature set size through a novel compression function.
• Enhanced SVM classification accuracy with a dramatically reduced input feature vector size.

Conclusions:

• The developed algorithm achieves high accuracy and efficiency, suitable for real-time epilepsy detection.
• High sensitivity and specificity were observed, with potential for further improvement in Generalized Tonic Clonic Seizures (GTCS).
• Future work includes optimizing the compression and SVM evaluation stages and acquiring more GTCS data.