Publication

Abstract : Every year, about 50 million people is detected with cardiovascular diseases, heart disease is foremost the most dangerous in healthcare industry, and any upcoming or new advancements help in increasing patient life. An electrocardiogram (ECG) has been widely utilized for identifying cardiac problems due to its simplicity and non-invasive nature. Arrhythmia is a disorder causing an increase or decrease in the rate or rhythm of the heartbeat, which may occur sporadically. There are many research and developments which made to increase the efficiency of the existing ECG signal analysis using specific classification and monitoring of each patient's electrocardiogram. This project discusses a method based on transfer learning for heartbeat classification and arrhythmia detection by using PhysioNet’s MIT-BIH database (which can accurately classify if in accordance with the AAMI EC57 standard). There have been many sources and references using the MIT-BIH dataset, and the recent papers indicate scope for improvement, by using newer technology and using different analysis methods. Here, we use ResNet 50 for classification, and the method can make predictions with an average accuracy of 99.96%. Furthermore, we suggest a method for converting one-dimensional time signals to two-dimensional images using a main angular matrix and then passing the images through the Gramian angular matrix and then passing images through transfer learning through ResNet 50 model to get an output of 98.86%. This will help clinicians with an alternative method to detect and differentiate between different types of arrhythmias with better accuracy.