Seizure detection using regression tree based feature selection and polynomial SVM classification.

Area of Science:

• Biomedical Engineering
• Computational Neuroscience
• Medical Signal Processing

Background:

• Epilepsy affects millions globally, necessitating accurate and efficient seizure detection methods.
• Current seizure detection systems often face challenges with hardware complexity and power consumption, limiting their real-world application.
• Patient-specific algorithms are crucial for improving the accuracy and reliability of seizure detection.

Purpose of the Study:

• To develop a novel patient-specific algorithm for detecting seizures in epileptic patients.
• To achieve low hardware complexity and low power consumption for practical implementation.
• To evaluate the algorithm's performance using intracranial electroencephalogram (iEEG) data.

Main Methods:

• Computation of spectrogram from fragmented one-second EEG signal clips using three or four electrodes.
• Extraction of spectral powers and spectral ratios as features.
• Feature selection using regression trees, followed by classification with a polynomial Support Vector Machine (SVM) of degree 2.

Main Results:

• Achieved 100.0% sensitivity and 99.9% specificity using half of the training data.
• Obtained an average Area Under Curve (AUC) of 0.9818.
• Demonstrated a mean detection horizon of 5.8 seconds and a mean AUC of 0.9136 for seizure detection on testing data.

Conclusions:

• The proposed patient-specific algorithm offers a highly accurate and efficient solution for seizure detection in epileptic patients.
• The algorithm's low hardware complexity and power consumption make it suitable for real-time, wearable applications.
• The method shows significant potential for improving the management and monitoring of epilepsy.