Reduction of false alarms in the intensive care unit using an optimized machine learning based approach

Area of Science:

• Biomedical Engineering
• Medical Informatics
• Artificial Intelligence in Medicine

Background:

• Intensive care units (ICUs) are burdened by a high rate of false alarms from bedside monitors, leading to alarm fatigue.
• A majority of current cardiac arrhythmia alarms are false, necessitating improved detection algorithms.
• Reducing false alarms is critical for patient safety and efficient clinical workflow in critical care settings.

Purpose of the Study:

• To develop and validate an optimized machine learning approach for reducing false arrhythmia alarms in the ICU.
• To improve the accuracy and efficiency of real-time cardiac monitoring systems.
• To distinguish true arrhythmic events from noise and artifacts in physiological signals.

Main Methods:

• Applied a three-stage methodology: signal processing for heartbeat annotation, feature extraction including signal quality indices (SQIs), and optimized machine learning.
• Utilized Random Forest algorithm with feature selection to enhance model efficiency and reduce complexity.
• Employed the dataset from the PhysioNet/Computing in Cardiology Challenge 2015 for training and testing.

Main Results:

• Achieved a score of 83.08 in the real-time category on the hidden test set, surpassing all previously published results.
• Successfully distinguished between normal physiological signals and noise/artifacts using extracted SQIs.
• Demonstrated improved performance in identifying true arrhythmic alarms.

Conclusions:

• The proposed signal processing and machine learning framework effectively reduces false alarms in ICU bedside monitoring.
• Optimized feature selection and SQIs are crucial for accurate arrhythmia detection and noise reduction.
• This approach represents a significant advancement in real-time cardiac monitoring and alarm management.