A fused electrocardiography arrhythmia detection method

Abstract

Recently, Electrocardiography (ECG) signals are commonly used in diagnosing the cardiac arrhythmia that shows up with the loss of the regular movement of the heart. Approximately 5% of the world population have cardio motor disorders. Therefore, usage of the ECG signals in biomedical signal processing algorithms and machine learning methods for automated diagnosis of this widespread health problem is a popular research topic. In this paper, the Particle Swarm Optimization (PSO) technique is implemented to tune the parameters of Tunable Q-Factor Wavelet Transform (TQWT) and the new generation feature generator Hamsi Hash Function (Hamsi-Pat) is used to obtain the characteristics of the signal. Sub-signals of 10 s obtained from the original ECG signal are divided into their sub-bands of 25 levels with PSO and TQWT. Each of these low pass filters generates 536 dimensional features by applying Hamsi-Pat and statistical methods. Then, all these features are combined and 536 × 25 = 13400-dimensional feature set is obtained. The features in the set are reduced and the best of them are selected by using the Iterative Neighborhood Component Analysis (INCA) method. Finally, the k-Nearest Neighbors (kNN) classification method is applied to the best features according to the City Block measurement criterion. All studies cited to compare the results in this paper also use the MIT-BIH Arrhythmia ECG database. Hence, the difference could be observed in the used techniques. In contrast to the existing studies, this study shows its superior performance by classifying all 17 classes simultaneously by applying a “fused” approach. The method in the paper reached 98.5% classification accuracy on the 17 classes of the MIT-BIH Arrhythmia ECG database. The results indicate that the proposed method showed better rates from the existing studies related to arrhythmia diagnosis using ECG signals in the literature.