Development of a model for automated detection of abnormalities in ECG

Abstract

New research avenues in healthcare and life sciences have opened up as a result of the emergence of computing paradigms and the capacity to apply computational approaches. The heart is a critical organ of the body, clinical examinations and research on the heart involve a large number of researchers. The use of an electrocardiogram to assess the electrical activity of the heartbeat has been investigated by researchers (ECG). Since then, the electrocardiogram (ECG) has become one of the most important and key diagnostic tools for detecting cardiac function issues. According to WHO data, cardiovascular disorders are currently the leading cause of death (CVDs). According to their figures, CVDs are responsible for nearly 30% of all deaths. The availability of ECG data, as well as the capability of using deep learning models and their robustness, has prompted academics to try to use them to improve ECG analysis accuracy. Using a 2D Convolutional Neural Network, we proposed an effective arrhythmia classification model in this paper (CNN). This research is centered on CNN optimization employing approaches such as batch normalization and data augmentation. This is an unsupervised learning-based technique in which we have avoided traditional data pre-processing and feature extraction while still achieving accuracy close to 90% and sensitivity near to the same level. We also offer a method for classifying heartbeats that is compliant with the AAMI EC57 guidelines and aids in the classification of five different types of arrhythmias. We were able to attain an average accuracy of closer to 97 percent using this strategy, which is based on the t-sne model. newline