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

Abstract

Epilepsy is often associated with modifications in autonomic nervous system, which usually precede the onset of seizures of several minutes. Thus, there is a great interest in identifying these modifications enough time in advance to prevent a dangerous effect and to intervene. In addition, these changes can be a risk factor for epileptic patients and can increase the possibility of death. Notably autonomic changes associated to seizures are highly depended of seizure type, localization and lateralization. The aim of this study was to develop a patient-specific approach to predict seizures using electrocardiogram (ECG) features. Specifically, from the RR series, both time and frequency variables and features obtained by the recurrence quantification analysis were used. The algorithm was applied in a dataset of 15 patients with 38 different types of seizures. A feature selection step, was used to identify those features that were more significant in discriminating preictal and interictal phases. A preictal interval of 15 minutes was selected. A support vector machine (SVM) classifier was then built to classify preictal and interictal phases. First, a classifier was set up to classify preictal and interictal segments of each patient and an average sensibility of 89.06% was obtained, with a number of false positive per hour (FP/h) of 0.41. Then, in those patients who had at least 3 seizures, a double-cross-validation approach was used to predict unseen seizures on the basis of a training on previous ones. The results were quite variable according to seizure type, achieving the best performance in patients with more stereotypical seizure. The results of the proposed approach show that it is feasible to predict seizure in advance, considering patient-specific, and possible seizure specific, characteristics.

Predictors of Long-Term Neurodevelopmental Outcome of Hypoxic-Ischemic Encephalopathy Treated with Therapeutic Hypothermia

“…HRV metrics can include both linear measures in time and frequency domains, and nonlinear metrics derived from complexity analysis. 95 HRV in the high-frequency (HF) range is attributed to parasympathetic activity, while low-frequency (LF) range is attributed to mainly sympathetic and parasympathetic activity. The LF/HF ratio serves as an indicator of sympathovagal balance.…”

Forecasting cycles of seizure likelihood

“…For instance, self‐reported stress is predictive of seizures 27,28 . Heart rate 29,30 and other physiological signals monitored from a smartwatch device 31 have also been used to forecast seizure likelihood. It is possible that these noncerebral biomarkers of seizure likelihood are useful because the same fundamental rhythms that modulate many aspects of human physiology also drive seizure risk.…”

The role of devices in managing risk

“…Whilst EEG recordings could be considered the optimum pre-ictal information, data acquisition is challenging and, on this basis, indirect or surrogate biomarkers have been evaluated. A pre-ictal increase in heart rate has been demonstrated in 36% of patients with temporal lobe epilepsy [14] and changes in heart rate variability have been identified in up to 70% of patients, 13.7 minutes before clinical seizure onset [15]. With recent technological advances in wearable devices monitoring cardiac function and other physiological parameters such as oxygen saturation it may be possible in future to provide patients with real-time seizure risk information with similar accuracy to the intracranial EEG recordings.…”

“…Summary of seizure detection devicesVariable sensitivity, nocturnal seizures only, high false alarm rate during periods of wakefulness. 70% (pre-ictal)[14,15] to 98% (ictal)[18] VNS (CBSDA) detection 66-74%[21,22] Stimulation: 35-41%[21, 22] All Suitable for most seizure types, can provide preictal warning. With VNS can also intervene Most devices have obtrusive electrodes or require VNS implantation.…”