Arrhythmia Classification Using EFFICIENTNET-V2 with 2-D Scalogram Image Representation

Cardiovascular disease is part of global death's main cause. It is the term for all types of diseases that affect the heart or blood vessels. Heart disease is a type of cardiovascular disease. It can be detected early by examining the arrhythmia presence. Arrhythmia is an abnormal heart rhythm that is commonly diagnosed and evaluated by analyzing electrocardiogram (ECG) signals. In classical techniques, a cardiologist/clinician used an electrocardiogram (ECG) to monitor the heart rate and rhythm of patients then read the journal activity of patients to diagnose the presence of arrhythmias and to develop appropriate treatment plans. However, The classical techniques take time and effort. The development of arrhythmias diagnosis, toward computational processes, such as arrhythmias detection and classification by using machine learning and deep learning. A convolutional neural network (CNN) is a popular method used to classify arrhythmia. Dataset pre-processing was also considered to achieve the best performance models. MIT-BIH Arrhythmia Database was used as our dataset. Our study used the EfficientN et-V2 which is a type of convolutional neural network to perform the classification of five types of arrhythmias. In pre-processing, the ECG signal was cut each 1 second (360 data), signal augmentation is applied to balance the amount of data in each class, and then the Continues Wavelet Transform (CWT) is employed to transform the ECG signal into a scalogram. The dataset is then distributed into subsets by using modulo operation to get variants of data in each subset. The colormap is applied to convert scalograms into RGB images. By this scheme, our study achieved superior accuracy than the existing method, with an accuracy rate of 99.97%.