Deep Learning based Cardiac Arrest Prediction using Vital sign and Lab code

Title : Deep Learning based Cardiac Arrest Prediction using Vital sign and Lab code
Published in : The 12th International Conference on Internet (ICONI 2020)
Author : Minsu Chae, Sangwook Han, Geonil Yun, Hyo-Wook Gil, Min Hong, DoKyeong Lee, HwaMin Lee
Corresponding author : HwaMin Lee
Location : Jeju ShinhwaWorld, Jeju, Korea

Abstract
There are patients in hospitals who are hospitalized for a variety of reasons. We conducted a study on predicting cardiac arrest on patients at Soonchunhyang University Cheonan Hospital. We studied patients from 2016 to 2019. We used deep learning via the LSTM model and the GRU model. We check density of each input feature according to cardiac arrest. We compared only patients with vital signs and lab data. We removed DBP, SBP, BodyTemperature, AST, ALT, WBC, Creatinine, and Bilirubin variable density because there was little difference between those with cardiac arrest and other patients. We experimented with the LSTM model and GRU Model. In this paper, deep learning-based cardiac arrest prediction using vital signs and lab data has high precision. In particular, when using the GRU model, even if the cardiac arrest of 0 to 24 hours for each record is changed, it has a sensitivity of 60%.

Acknowledgments
This research was supported by the Bio & Medical Technology Development Program of the National Research Foundation (NRF) funded by the Korean government (MSIT) (No. NRF-2019M3E5D1A02069073) and supported by the MSIT(Ministry of Science and ICT), Korea, under the ITRC(Information Technology Research Center) support program(IITP-2020-2015-0-00403) supervised by the IITP(Institute for Information & communications Technology Planning & Evaluation)