Abstract
BACKGROUND: Timely detection is crucial for managing cardiovascular diseases. Recently developed computational tools to analyze RR interval (RRI) sequences offer cost-effective means for early cardiac screening and monitoring with consumer-grade heart rate devices. OBJECTIVE: The purpose of this study was to demonstrate detection of congestive heart failure (CHF) from RRIs by discriminating CHF from both healthy controls and patients with present atrial fibrillation (AF). We also examined the detection's consistency regarding CHF severity and AF episode frequency. METHODS: We analyzed RRIs extracted from several datasets of long-term electrocardiographic (ECG) recordings. We use detrended fluctuation analysis (DFA) to evaluate the correlations of RRI, that is, how changes in the RRIs affect changes at another time. Furthermore, we utilized dynamical detrended fluctuation analysis (DDFA), which provides further insights into how the correlations change over time and different time scales. The resulting (D)DFA scaling exponents are used as features in classification, distinguishing CHF, AF, and healthy controls using the XGboost ensemble learning technique. RESULTS: Our (D)DFA computations revealed distinct RRI characteristics for CHF and AF patients during long-term ECG recordings, aiding disease detection. The DDFA-based classification pipeline detects CHF/AF from healthy controls with 90% sensitivity and 92% specificity. The 3-class classification algorithm correctly detects 78% of AF cases, 78% of CHF cases, and 91% of healthy cases. The DDFA results show consistency regarding CHF severity and AF episode frequency. CONCLUSION: We achieved high confidence in detecting CHF, with DDFA showing excellent classification accuracy, especially in multiclass tasks. This approach highlights the potential of noninvasive, cost-efficient RRI analysis for early detection of CHF and AF.