Open Access

ARTICLE

Arrhythmia Detection and Classification by Using Modified Recurrent Neural Network

Ajina Mohamed Ameer^*, M. Victor Jose

Department of CSE, Noorul Islam Centre for Higher Education, Kumaracoil, 629180, India

* Corresponding Author: Ajina Mohamed Ameer. Email:

Intelligent Automation & Soft Computing 2022, 33(3), 1349-1361. https://doi.org/10.32604/iasc.2022.023924

Received 27 August 2021; Accepted 08 December 2021; Issue published 24 March 2022

Abstract

This paper presents a novel approach for arrhythmia detection and classification using modified recurrent neural network. In medicine and analytics, arrhythmia detections is a hot topic, specifically when it comes to cardiac identification. In the research methodology, there are 4 main steps. Acquisition and pre-processing of data, electrocardiogram (ECG) feature extraction utilizing QRS (Quick Response Systems) peak, and ECG signal classification using a Modified Recurrent Neural Network (Modified RNN) for arrhythmia diagnosis. The Massachusetts Institute of Technology-Beth Israel Hospital. (MIT-BIH) Arrhythmia database was used, as well as the image accuracy. Medium filter is used in the pre-processing. Feature extraction is done with morphological and dynamic features to detect morphological arrhythmia the shape morphological properties of the ECG signal. Dynamic arrhythmia could be diagnosed by having some feature of the ECG signal such as amplitude and position of the QRS peak. Using the modified pan Tompkins algorithm, arrhythmia was detected. Load the ECG signal after getting QRS complex R and p peak of ECG signal detected. For the deep learning classification modified RNN is used as a classifier. The modified RNN is trained independently for each of the 17 classes using training and validation data, Data from the validation phase is utilized to calculate network parameters tweaking. The acquired results demonstrate the proposed method’s effectiveness. The overall classification accuracy for 17 cardiac arrhythmias was 95.33%. For each 10 s ECG data, the classification time was 0.015 s. The proposed technique is compared in terms of accuracy to that of other existing techniques, revealing that the new method outperforms them.

---

Keywords

---