Online apnea-bradycardia detection using recursive order estimation for auto-regressive models

This study aims to detect apnea-bradycardia (AB) episodes from preterm newborns, based on the analysis of electrocardiographic signals (ECG). We propose the use of an auto-regressive (AR) model with undetermined orders to capture all possible linear dependency of the RR interval time series extracted from ECG. An on-line algorithm inspired from the Kalman filtering technique is designed to follow the evolution of the AR model's order distribution. The detection sensitivity (TP/(TP + FN)) reaches 91:5% over a total of 50 episodes with perfect specificity (TN/(FP+TN)=100%). From the clinical point of view, it is essential to achieve reliable early stage detection of AB episodes to enable the initiation of quick nursing actions. Our proposed method achieves a delay of 5.08s ± 2.90 compared with the experts' off-line annotations, knowing that the mean intervention time (duration from the generation of the alarm to the initiation of manual stimulation) is reported to be 33 seconds from a recent study [5].