Computationally Efficient Algorithm for Atrial Fibrillation Detection using Linear and Geometric Features of RR Time-Series Derivatives

This paper proposes a computationally efficient algorithm for Atrial Fibrillation (AF) detection using linear and geometric features of multiple RR time series derivatives. Our main goal is to demonstrate that the detection of AF episodes can be improved using complementary heart-rate dynamics over traditional approaches. Herein, we investigate the combination of eleven dynamic forms, namely the RR interval time series, the first five standards, and absolute derivatives. From each dynamic, we extract 11 features, yielding a set of 121 components. Therefore, we applied an ANOVA test and a recursive feature elimination strategy to eliminate uninformative and irrelevant features and construct an appropriate subset of relevant features. Next, we perform multilayer perceptron (MLP) for model building. The process evaluates and selects the most accurate model based on sensitivity, specificity, and accuracy performance metrics. The results highlight the strengths of the proposed approach, which could serve as valuable decision support for AF diagnosis.