A Novel Severity Ranking Approach for Continuous Monitoring of Heart Disease Progression Using Beat-wise Classification of ECG

Irregularities in heartbeats and cardiac functioning outside of clinical settings are often not available to the clinicians, and thus ignored. But monitoring these with high-risk population might assist in early detection and continuous monitoring of Atrial Fibrillation (AF). Wearable devices like smart watches and wristbands, which can collect electrocardiograph (ECG) signals, can monitor and warn users of unusual signs in a timely manner. Thus, there is need to develop a real-time monitoring system for AF from ECG. We propose an algorithm for a simple beat-by-beat ECG signal multilevel classifier for AF detection and a quantitative severity scale (0 to 1) for user feedback. For this study, we used ECG recordings from MIT BIH Atrial Fibrillation, MIT-BIH Long-term Atrial Fibrillation Database. All ECG signals are preprocessed for reducing noise using filter. Preprocessed signal is analyzed for extracting 39 features including 20 of amplitude type and 19 of interval type. The feature space of all ECG recordings is considered for classification. Training and testing data include all classes of data i.e., beats to identify various episodes for severity. Feature space from the test data is fed to the classifier which determines the class label based on trained model. A class label is determined based on number of occurrences of AF and other Arrhythmia episodes, such as AB (Atrial Bigeminy), SBR (Sinus Bradycardia), SVTA (Supra ventricular Tachyarrhythmia). Accuracy of 96.7764% is attained with Random Forest algorithm. Furthermore, precision and recall are determined based on correct and incorrect classifications for each class. Precision and Recall on average for Random Forest Classifier are obtained as 0.968 and 0.968 respectively. This work provides a novel approach to enhance existing method of AF detection by identifying heartbeat class and calculates a quantitative severity metric that might help in early detection and continuous monitoring of AF.