Machine Learning Analysis of Heart Rate Variability for the Detection of Seizures in Comatose Cardiac Arrest Survivors

Objective: Heart rate variability (HRV) reflects autonomous nervous system disturbance and is used for seizure prediction. The aim of this study was to develop a real-time, continuous physiological medical signal data acquisition system in seizure detection for intensive care unit (ICU). Methods: This prospective study was conducted in National Taiwan University Hospital from August 2018 to October 2019. This study included 20 patients who (a) had a sustained return of spontaneous circulation following out-of-hospital cardiac arrest, (b) were over 18 years old, (c) and were admitted to the emergency ICU for post-cardiac-arrest care. One-lead electrocardiography and bilateral two-channel electroencephalography recorders were synchronically used to conduct measurements for a maximum of 72 hours. The recorded data were wirelessly real-time transmitted by a proxy transmitting module through an access point and a local gateway. A system with a novel algorithm processed the signals and conducted feature extraction and supervised learning for seizure detection. Results: A total of 89 nonseizure and 83 seizure events were detected by the system. Seizure occurred in two-thirds of the patients assessed by intensivists and neurologists. Four HRV parameters, namely standard deviation of normal-to-normal R-wave intervals, high frequency, low frequency–high frequency ratio, and sample entropy, were determined as potential features for identifying seizures. The sensitivity and specificity of the developed system were 0.74 and 0.81, respectively, and the positive predictive value was 0.82. Conclusion: The developed system can be used to identify seizure events through HRV features. Significance: The current study achieved real-time seizure detection and overcame previous limitations on continuity and accessibility.