Automatic ECG Arrhythmia Recognition using ANN and CNN

Present research highlights the need for more patient-oriented monitoring systems for cardiac health, especially in the aftermath of COVID-19. The study introduces a contactless and affordable ECG device capable of recording heart arrhythmias for remote monitoring, which is vital in managing the rising incidence of untimely heart attacks. Two deep learning algorithms have been developed to design the system: RCANN (Real-time Compressed Artificial Neural Network) and RCCNN (Real-time Compressed Convolutional Neural Network), respectively, based on ANN and CNN. These methods are designed to classify and analyze three different forms of ECG datasets: raw, filtere and filtered + compressed signals. These were developed in this study to identify the most suitable type of dataset that can be utilized for regular/remote monitoring. This data is prepared using online ECG signals from Physionet(ONLINE) and the developed real-time signals from Arduino ECG sensor device. Performance is analysed on the basis of accuracy, sensitivity, specificity and F1 score for all kinds of designed ECG databases using both RCCNN and RCANN. For raw data, accuracy is 99.2%, sensitivity is 99.7%, specificity is 99.2%, and F1-Score is 99.2%. For RCCNN, accuracy is 93.2%, sensitivity is 91.5%, specificity is 95.1%, and F1-Score is 93.5% for RCANN. For Filtered Data, accuracy is 97.7%, sensitivity is 95.9%, specificity is 99.4%, and F1-Score is 97.6%. For RCCNN, accuracy is 90.5%, sensitivity is 85.8%, specificity is 96.4%, and F1-Score is 90.9% for RCANN. For Filtered + compressed data, accuracy is 96.6%, sensitivity is 97.6%, specificity is 95.7%, and F1-Score is 96.5%. For RCCNN, accuracy is 85.2%, sensitivity is 79.2%, specificity is 94.5%, and F1-Score is 86.7% for RCANN. The performance evaluation shows that RCCNN with filtered and compressed datasets outperforms other approaches for telemonitoring and makes it a promising approach for individualized cardiac health management.