Rapid Detection of Heart Rate Fragmentation and Cardiac Arrhythmias: Cycle-by-Cycle rr Analysis, Supervised Machine Learning Model and Novel Insights

Abstract

Heart rate dynamics are a macroscopic indicator of cardiac health. Sino-atrial degradation manifested as heart rate fragmentation (HRF) are analyzed using rr values (relative-RR intervals) derived from the inter-beat-intervals of ECGs. The rr-value is useful for the analysis of cycle-by-cycle variations such as HRF and arrhythmias. Three novel metrics developed in this work: CM20, Z3e20 and sPIP, along with two conventional metrics: SDNN and LFHF ratio are used for the detection of HRF and arrhythmias. The supervised machine learning technique of random forests is applied to develop the classification model. For this, we used a balanced dataset of 300 cases comprising of arrhythmic, non-arrhythmic coronary artery disease, and individuals without any medically significant cardiac conditions. The model was tested on 104 independent cases. The F1 score of the classifier is 91.1% without any adjustments for age, gender, prior medical conditions, etc. Insight into threshold values of heart rate dynamics for arrhythmic, heart rate fragmentation and normal cases are obtained from a single decision tree model.