Early Detection of True or False Cardiac Alarms in the ICU
Conference Year
January 2021
Abstract
Early detection of whether a cardiac alarm is true or false is as critical as accurate detection in intensive care. Delayed detection may lead to a patient’s death if the alarm is true or to disruption if false. There has been a body of research, as signified by the 2015 PhysioNet/CinC Challenge, and due accomplishments have been made in the relevant computational technology. The focus of the research, however, has been on the accuracy of false-alarm detection, and yet the highest accuracy known thus far is in the lower 80% range. Our work achieves much higher accuracy by utilizing state-of-the-art machine learning methods and, more importantly, achieves very early detection, almost at the onset of a cardiac alarm. This is enabled by the machine learning method used, which is a combination of ResNet and BiLSTM. Using the PhysioNet dataset of 750recorded ECG segments published with the challenge, our method achieved 96% accurate false alarm detection in0.51 seconds on average over all segments.
Primary Faculty Mentor Name
Byung Lee
Status
Undergraduate
Student College
College of Engineering and Mathematical Sciences
Program/Major
Biomedical Engineering
Second Program/Major
Computer Science
Primary Research Category
Engineering & Physical Sciences
Secondary Research Category
Biological Sciences
Early Detection of True or False Cardiac Alarms in the ICU
Early detection of whether a cardiac alarm is true or false is as critical as accurate detection in intensive care. Delayed detection may lead to a patient’s death if the alarm is true or to disruption if false. There has been a body of research, as signified by the 2015 PhysioNet/CinC Challenge, and due accomplishments have been made in the relevant computational technology. The focus of the research, however, has been on the accuracy of false-alarm detection, and yet the highest accuracy known thus far is in the lower 80% range. Our work achieves much higher accuracy by utilizing state-of-the-art machine learning methods and, more importantly, achieves very early detection, almost at the onset of a cardiac alarm. This is enabled by the machine learning method used, which is a combination of ResNet and BiLSTM. Using the PhysioNet dataset of 750recorded ECG segments published with the challenge, our method achieved 96% accurate false alarm detection in0.51 seconds on average over all segments.