Classify arrhythmia by using 2D spectral images and deep neural network

Electrocardiogram (ECG) is the most common method for monitoring the working of the heart. ECG signal is the basis to determine normal or abnormal rhythm, thereby helping to accurately diagnose cardiovascular diseases. Therefore, an automatic algorithm to detect and diagnose abnormal heart rhythms is essential. There are many methods of classifying arrhythmias using machine learning algorithms such as k-nearest neighbors (KNN), support vector machines (SVM), based on the features extracted from the record of ECG signal. Actually, deep learning algorithms are evolving and highly effective in image analysis and processing. In this research, a dense neural network model is proposed to classify normal and abnormal beats. Input ECG signal presenting a time series is converted into 2-D spectral image by applying wavelet transform. Our research is evaluated based on using the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database. The accuracy of the classification algorithm we employ is 99.8%, demonstrating the model's validity when compared to other reports' findings. This is the foundation for our algorithm to prove it can be utilized as an efficient model for categorizing arrhythmia using ECG signals.