Multi-modal integrated approach towards reducing false arrhythmia alarms during continuous patient monitoring: The Physionet Challenge 2015

This work presents a solution for the Physionet Challenge 2015 regarding false alarm reduction in ICU. False alarms can result in alarm fatigue, i.e. reduced responsiveness of the ICU personnel to the true alarms due to an enormous number of false alarms. As a result, it is necessary to effectively suppress the false alarms while ensuring that the true alarms are not ignored. The challenge data contains five different types of alarms which are treated as independent problems in this paper. A separate subroutine is used for each alarm which is composed of two stages, peak detection and alarm verification. This paper uses a multi-modal peak detection algorithm that uses the information from all the available signals and combines the results from several peak detection algorithms to create a robust peak detection algorithm. The alarm verification stage is alarm dependent, composed of simple decision criteria or a complicated neural network model. The proposed approach achieves an overall score of 74.48 for the real-time event, where only the portions of the signals prior to the alarm are utilized, and 76.57 for the retrospective event, where 30 seconds of the signals after the alarm are used as well.