Seizure detection method based on fractal dimension and gradient boosting.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Long-term electroencephalography (EEG) monitoring is vital for epilepsy management.
• Accurate and automated seizure detection is a critical need for clinical practice.
• Existing methods often lack patient-specific adaptability.

Purpose of the Study:

• To develop and evaluate a patient-specific automated method for epileptic seizure detection using EEG.
• To improve the accuracy and reliability of seizure detection in long-term monitoring.
• To provide a robust tool for analyzing EEG data in epilepsy patients.

Main Methods:

• Preprocessing of multichannel EEG data.
• Estimation of fractal dimensions using a k-nearest neighbor algorithm.
• Feature vector construction and classification using a gradient boosting model.
• Postprocessing steps including smoothing, thresholding, and temporal union operations.
• Patient-specific model training and validation on the Freiburg dataset.

Main Results:

• Achieved an average epoch-based sensitivity of 91.01% and specificity of 95.77%.
• Obtained an average event-based sensitivity of 94.05% with a low false detection rate of 0.27/h.
• Demonstrated effective performance on EEG data from 21 epilepsy patients.

Conclusions:

• The proposed patient-specific method offers a promising approach for automated epileptic seizure detection.
• The high sensitivity and specificity indicate the clinical utility of this technology.
• This method can aid in more efficient and accurate long-term EEG monitoring for epilepsy.