Automatic Detection of Cardiac Arrhythmia Using Deep Transfer Learning based Approach on ECG

With the frequency of cardiovascular issues rising, the automated categorization analysis of electrocardiogram signals (ECG) has become important in the medical assessment of disorders. A convolutional neural network (CNN)-based method is recommended for classifying ECG signals. Convolutional neural networks (CNNs) used in the deep learning technique may automatically detect ECG data without the need for simple manual feature extractions, but they need a lot of attention. In this work, AlexNet and ResNet-18, two pre-trained deep learning networks, were utilized to identify and detect cardiac arrhythmia, and their performance was evaluated across a number of metrics. There were some modifications within the layers of the two proposed CNN models for classifying cardiac arrhythmias, and spectrograms were employed as the model's input images. The practical findings demonstrate that the suggested method outperforms various well-used ECG classification methods, achieving promising classification accuracy, sensitivity, and precision of greater than 99%. Although, along with different learning rates and other modifications, the epoch sizes and minimum batch sizes were hyperparameters that were adjusted, the results were still impressive.