AED Automatic Diagnose Algorithm Based On Hurst Exponent

Automated External Defibrillator(AED) is a first-aid equipment that significantly improves the survival chances of patients suffer from fatal cardiac arrest(ventricular fibrillation and ventricular tachycardia) out-of-hospital. According to the recommendations of American Heart Association(AHA), the arrhythmia analysis algorithm used by AED should produce a sensibility greater than 90% for shockable rhythms and a specificity greater than 95% for non-shockable rhythms.Hurst exponent, a non-linear descriptor that differentiates time series data by the degree of autocorrelations, can be a reasonable candidate for the auto diagnose algorithm of AED, as the ECG signal is by nature a time series that contains autocorrelations and self-similarities.We propose a time series analytical method that leverages Hurst exponent to quantitate the non-linear dynamical features of ECG signals. By setting up a proper threshold, the ECG signals in question are classifiable based on Hurst exponent quantitations, and diagnose of their underlying pathological conditions is also feasible. To verify the reliability and performance of this method, ECG signals from MIT-BIH Arrhythmia DB, CUDB and VFDB are used. The result shows a sensibility of 96.27% and a specificity of 96.97%, which satisfies the requirements of AHA.The above result is achieved with window length set to 3 seconds, which is much shorter than the 8 seconds window required by many arrhythmia analysis algorithm reported. In practice, this shorter-window-length feature might be applicable to enhance the real-time capability of AED auto diagnose.