Compact 1D Self-Operational Neural Networks with Feature Injection for Global ECG Classification

for solving problems in ECG classification readings for rhythm detection, a novel deep learning approach is presented in this paper. To capture important information about heart cycles like morphology and timing, a proposed technique integrates feature injection with a compact 1D Self-Operational Neural Network (Self-ONN). This allows for fluent and engaging data recording. With superior performance, the network surpasses current models in classification while maintaining minimal computational complexity. Its outstanding results are achieved through training with the MIT-BIH arrhythmias benchmark database. The network displays exceptional levels of accuracy, recall, and F1-score for normal segments (99%), supraventricular ectopic beats (82%), and ventricular ectopic beats (94%), demonstrating its remarkable capabilities. Here study demonstrates the applicability of the approach in wearable ECG sensors for self-monitoring and early detection of cardiovascular diseases. The paper offers comprehensive insights into network architecture, data processing stages, as well as comparisons with current algorithms. The suggested approach outperforms others in terms of F1-score, accuracy, sensitivity, specificity, and positive predictive value when applied to various beat types. The report contains a computational complex analysis that emphasizes the compactness of the model and its importance in reducing the difference in performance between patient-specific and global ECG classification techniques.