Reducing false arrhythmia alarm rates using robust heart rate estimation and cost-sensitive support vector machines.

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Informatics

Background:

• False critical arrhythmia alarms in patient monitoring systems increase healthcare costs and reduce clinician trust.
• Existing alarm systems often struggle with signal noise and complex arrhythmia patterns, leading to missed detections or false alarms.

Purpose of the Study:

• To develop and evaluate a novel system for reducing false critical arrhythmia alarms using advanced signal processing and machine learning techniques.
• To improve the accuracy of arrhythmia detection by employing robust heart rate estimation and cost-sensitive classification algorithms.

Main Methods:

• Utilized multimodal physiological data (ECG, blood pressure, photoplethysmogram) from PhysioNet databases for training and testing.
• Implemented a signal quality modified Kalman filter for robust heart rate estimation.
• Extracted heart rate variability and statistical ECG features, optimized using a genetic algorithm (GA).
• Employed cost-sensitive support vector machines (CSSVMs) for classifying alarms, prioritizing the high cost of false negatives.

Main Results:

• The system demonstrated a high true positive rate of 95% for critical arrhythmia detection.
• Achieved a true negative rate of 85%, indicating effective reduction of false alarms.
• The combination of robust heart rate estimation, feature selection via GA, and CSSVMs proved effective in enhancing alarm accuracy.

Conclusions:

• The proposed method significantly reduces false critical arrhythmia alarms, enhancing patient safety and clinical workflow efficiency.
• Cost-sensitive machine learning approaches are crucial for medical alarm systems where misclassification has high stakes.
• This approach offers a promising solution for improving the reliability of automated cardiac monitoring systems.