Poster: DeePTOP: Personalized Tachycardia Onset Prediction Using Bi-directional LSTM in Wearable Embedded Systems.

Monitoring tachycardia and early intervention can reduce the occurrence of heart failure, cardiac arrest, death, etc. In this paper, we propose a novel personalized Bidirectional Long Short-Term Memory (BLSTM) model for early individualized tachycardia diagnosis. It leverages continuous monitored vital sign including heart rate (HR), respiratory rate (RR), and blood oxygen saturation (SpO2), which are feasibly acquired by wearable embedded systems, as well as the admission information. The Area Under the Curve (AUC) of our model in our clinical experiments achieves 0.82 when predicting the onset of tachycardia 6 hours (6h) in advance, which precedes several baseline models.