Real-time detection of atrial fibrillation using a low-power ECG monitor

A study was performed to determine the feasibility of a miniature, low-power ECG monitor capable of real time, automatic detection of atrial fibrillation. An original arrhythmia detection scheme was devised and tested using the MIT arrhythmia data available on PhysioNet. Five beat and five rhythm detectors were constructed and the regression values of each were passed onto two further classifiers for ultimate detection of atrial fibrillation. Tests showed that normal sinus rhythm could be detected with 93.06% sensitivity and 95.08% specificity and atrial fibrillation with 94.76% sensitivity and 92.48% specificity. The target device was constructed and fast, efficient algorithms were developed to carry out the signal processing and classification processes. Power consumption was measured at 30mW giving 96 hours of continuous operation. The computation time for the signal sub-band filtering and heart beat interval calculations was measured at 2.1ms per 8ms interval, and heart beat classification at 10.2ms per classifier per beat detected. This research demonstrates that the design of a low-powered, low-cost, miniature ECG monitor having the ability to automatically detect atrial fibrillation in real time is feasible.