Reducing false arrhythmia alarms in the ICU by Hilbert QRS detection

In this study, we develop algorithms that reduce the arrhythmia false alarms in the ICU by processing the four signals of Photoplethysmography (PPG), arterial blood pressure (ABP), ECG Lead II, and Augmented right arm ECG. Our algorithms detect five arrhythmias including asystole, extreme bradycardia, extreme tachycardia, ventricular tachycardia (VT), and ventricular flutter or fibrillation (VF). Real time algorithm is provided. Our processing proceeded as follows. Firstly, preprocessing was applied to the ECG signals by two median filters in order to remove the baseline wander and high-frequency noise. Then a Hilbert-transform based QRS detector algorithm was used to detect R waves from the ECG signals. Following this, RR intervals were calculated from the available ECG signals. Pulse onset points of the pulsatile signals (PPG and ABP) were also detected and the signal quality index (SQI) of the four signals was measured. The ECG based RR intervals were combined with the pulsatile signal based RR intervals using the algorithms provided by the CinC2015 competition organizers. The combined RR intervals were thresholded at the clinically important values for the five arrhythmias. Template matching was used to detect ventricular tachycardia (VT) and power spectrum of ECG signals and identifying the VF frequency components employed to investigate ventricular fibrillation. Our highest overall result was a 98% True Positive Rate (TPR), 66% True Negative Rate (TNR) with a score of 74.03% for the retrospective algorithm. For the realtime algorithm, we achieved a 98% TPR, 65% TNR and a score of 69.92%.