Implementation of Discrete Wavelet Transform and Xception for ECG Image Classification of Arrhythmic Heart Disease Patients

Abstract

The electrocardiogram (ECG) is one of the most important methods in the process of diagnosing heart disease. Visualizes the voltage and time relationship of the electrical activity of the heart. Cardiovascular or heart disease can be classified into several types, one of which is arrhythmia, a condition that involves changes in heartbeat rhythm, either too fast or too slow at rest. This study aims to develop a cardiac arrhythmia classification model using Deep Wavelet Transform (DWT) and Xception. It was evaluated on 2,200 spectrogram samples from the MIT-BIH dataset, containing normal and arrhythmia classes. The process compared epochs 30, 50, and 100 with learning rates of 0.001 and 0.0001 using cross-validation. Data were converted into spectrogram images for classification with Xception. The highest accuracy, 99.79%, was achieved at epoch 100 with a 0.0001 learning rate. Then, the highest precision occurs when the epoch is 50 with a learning rate of 0.001 and 0.0001, which is 100%. Lastly, Xception performed very well in the ECG image classification. This advantage demonstrates the ability of the model to recognize complex patterns in ECG data more effectively, increasing the reliability of arrhythmia detection. In addition, using DWT as a feature extraction technique allows better signal processing,
which contributes to optimal results.