Deep Learning for ECG-Based Arrhythmia Classification: A 1D-CNN with Optimization Techniques

This paper describes a novel deep-learning method derived from extended-duration electrocardiography. (ECG) data processing for the diagnosis of cardiac arrhythmia (5 classes). Given that over 50 million individuals world-wide are at risk of developing heart disease, preventing cardiovascular disease is one of the most crucial responsibilities of any healthcare system. Despite the widespread use of automated ECG signal processing, the techniques currently in use are inadequate. Our research aimed to develop a novel deep learning-based technique for the rapid and accurate classification of cardiac arrhythmias. Rather than using traditional techniques of handcrafted feature extraction and selection, a comprehensive end-to-end framework was created. The creation of a novel 1 dimensional Convolutional Neural Network (1D-CNN) model is our primary contribution. The suggested technique combines feature extraction and selection with classification in a single step, making it: 1) efficient; 2) quick and responsive 3) non-complex; and 4) easy to use. At a 97.81% recognition accuracy level across 5 cardiac arrhythmia diseases (classes), Deep 1D-CNN was able to classify data in 94 ms per sample. Our results are among the best to date when compared to the existing study, and our approach may be used with cloud computing and mobile devices.