Heartbeat classification using local transform pattern feature and hybrid neural fuzzy-logic system based on self-organizing map

Abstract

Abstract An automatic heartbeat classification method using electrocardiogram is important in assisting doctors and experts with the diagnosis of cardiac diseases. In this study, we introduce a novel algorithm based on a local transform pattern (LTP) with a hybrid neural fuzzy-logic system with a self-organizing map (NF). We extracted a histogram feature with multi-dimension using three feature extraction methods based on an LTP as a 1D local binary pattern, 1D local gradient pattern (1DLGP), and local neighbor descriptor pattern. The self-organizing map was then applied to a fuzzy-logic system for increasing the classification accuracy and reducing the time consumed. According to the recommendations for the advancement of medical instrumentation, we validated the proposed heartbeat classification method using five heartbeat classes—normal or bundle branch block, supraventricular ectopic, ventricular ectopic, fusion of ventricular and normal, and unknown beat. Experimental results show the performances of the proposed method using 1DLGP+NF with 196 (14 by 14) feature dimensions as robust performance at 87% (sensitivity), 73.8% (positive predictivity), 1.1% (false positive rate), and 98.84% (accuracy). We found that our study has a significant impact on heartbeat recognition methods, which are crucial functions in healthcare and medicine systems.