Patient-specific seizure prediction based on heart rate variability and recurrence quantification analysis

Area of Science:

• Neuroscience
• Biomedical Engineering
• Cardiology

Background:

• Epilepsy is linked to autonomic nervous system (ANS) changes preceding seizures.
• These ANS modifications can indicate seizure onset and pose risks, including mortality.
• Understanding seizure-specific autonomic patterns is crucial for prediction.

Purpose of the Study:

• To develop a personalized algorithm for predicting epileptic seizures using electrocardiogram (ECG) features.
• To identify significant ECG-derived features for distinguishing pre-seizure (preictal) and seizure-free (interictal) states.
• To evaluate the feasibility of a patient-specific seizure prediction model.

Main Methods:

• Utilized RR series from ECGs, analyzing time, frequency, and recurrence quantification analysis features.
• Employed a feature selection process to identify discriminative preictal and interictal markers.
• Developed a Support Vector Machine (SVM) classifier for phase classification, with a 15-minute preictal interval.
• Validated the model using patient-specific and cross-validation approaches.

Main Results:

• Achieved an average sensitivity of 89.06% with 0.41 false positives per hour in classifying preictal and interictal phases.
• Prediction performance varied by seizure type, with better results for more stereotypical seizures.
• Demonstrated the feasibility of predicting seizures using patient-specific characteristics.

Conclusions:

• Patient-specific ECG analysis can effectively predict epileptic seizures.
• The developed SVM model shows promise for early seizure detection.
• Further refinement based on seizure type could enhance prediction accuracy and clinical utility.