Wearable Epileptic Seizure Prediction System with Machine-Learning-Based Anomaly Detection of Heart Rate Variability

A warning prior to seizure onset can help improve the quality of life for epilepsy patients. The feasibility of a wearable system for predicting epileptic seizures using anomaly detection based on machine learning is evaluated. An original telemeter is developed for continuous measurement of R-R intervals derived from an electrocardiogram. A bespoke smartphone app calculates the indices of heart rate variability in real time from the R-R intervals, and the indices are monitored using multivariate statistical process control by the smartphone app. The proposed system was evaluated on seven epilepsy patients. The accuracy and reliability of the R-R interval measurement, which was examined in comparison with the reference electrocardiogram, showed sufficient performance for heart rate variability analysis. The results obtained using the proposed system were compared with those obtained using the existing video and electroencephalogram assessments; it was noted that the proposed method has a sensitivity of 85.7% in detecting heart rate variability change prior to seizures. The false positive rate of 0.62 times/h was not significantly different from the healthy controls. The prediction performance and practical advantages of portability and real-time operation are demonstrated in this study.