Heart Arrhythmia Classification Based on Statistical Moments and Structural Co-occurrence.

The electrocardiogram (ECG) is a widely disseminated method for detecting heart diseases due to its lower cost than other tests. But, some steps are important for detecting cardiac arrhythmias in ECG signals, which are: preprocessing, segmentation, feature extraction and classification. In this work, we assess how four non-morphological feature extraction methods provide useful ECG classification. Moreover, we propose an innovation in the configuration of the structural co-occurrence matrix (SCM), by combining it with the Fourier transform to extract the main frequencies of the signal. We tested theses methods on four well-known classifiers used in the literature and compare the results with six classical feature extraction methods. Moreover, we followed high standard protocols for developing expert systems for clinical usage. The database chosen for evaluation is the MIT-BIH arrhythmia database. We increased the identification of heart dysrhythmia by 2%, representing an advance on reports on the literature. The developed system is 1.3% more reliable than the current best approach reported, being 10 6 times faster, as well. The HOS with naive Bayes classified pathologies in 22 patients with 94.3% of accuracy. We perceived that SCM–Fourier is 1.5% more accurate than the SCM or Fourier standalone. The feature extractor proposed in this paper compress 97% of the useful information to provide a reliable arrhythmia classification. [ABSTRACT FROM AUTHOR]