Enhancing Epileptic Seizure Prediction Using Heart Rate Variability

Epileptic seizure prediction is a challenging problem in neurological monitoring, with important implications for patient safety and quality of life. While electroencephalography (EEG) is the standard modality for monitoring brain activity, its usability limitations motivate the exploration of alternative signals. In this context, a promising option is the electrocardiogram (ECG), which can be recorded using wearable devices, even over long periods. Single-lead ECG signals allow the estimation of heart rate variability (HRV), a marker of autonomic nervous system activity. However, HRV exhibits high inter-patient variability, making reliable pre-ictal pattern identification difficult. This thesis investigates ECG-derived HRV for patient-specific seizure prediction using single-lead ECG recordings from patients with drug-resistant epilepsy. Time-domain, frequency-domain, and nonlinear HRV features were extracted using a sliding-window approach. Feature-level pre-seizure dynamics were analyzed by comparing early baseline and late pre-seizure periods, and a patient-specific Multivariate Statistical Process Control (MSPC) framework was then employed to detect sustained deviations from baseline HRV behavior. No individual HRV feature showed a consistent group-level pre-seizure marker. The time-domain and compact multi-domain MSPC configurations both detected sustained candidate pre-seizure HRV deviations in 18 of 22 recordings, corresponding to a detection rate of 81.8%. For the compact multi-domain model, the mean warning time was 20.97 ± 6.48 minutes, with a range of 5.58–27.58 minutes before seizure onset. Although the two configurations achieved the same aggregate detection rate, they showed different patient-level behavior. These findings suggest that HRV does not provide a universal pre-ictal biomarker, but patient-specific multivariate monitoring can identify sustained autonomic deviations before seizure onset in a substantial proportion of recordings.