Arrhythmia Detection in Single-Lead Heartbeat Using ECG Residual Architecture

Cardiovascular diseases (CVD) stand as one of the gravest threats to human life. The electrocardiogram (ECG) emerges as a highly effective tool for CVD detection. These systems extract important information from a patient’s cardiac conditions and provide it to specialists. However, accurately and quickly diagnosing arrhythmia in ECG classification systems is challenging due to factors such as noise, individual variability in the morphology of heartbeats, and data imbalances. Achieving accurate and timely diagnoses of CVDs is vital for effective treatment and patient recovery. In this paper, we introduce a new CNN model designed for classifying heartbeat segments extracted from single-lead ECG. Our approach involves utilizing discrete wavelet transformation with the Sym7 mother wavelet and employing the SMOTE oversampling algorithm as a pre-processing step. The feature extraction and classification are performed by a 1D Residual Convolutional Neural Network. To assess the effectiveness of our proposed model, we conducted both training and testing phases using the MIT-BIH dataset, to identify five arrhythmia categories. We designed our experiment with two distinct scenarios: intra-patient and inter-patient. Our results demonstrate outstanding average classification accuracy of 99.53% for intra-patient and 97.87% for inter-patient, surpassing the performance results reported in recent literature on similar studies. Furthermore, our proposed model exhibits excellent performance in other evaluation metrics, including precision and sensitivity, indicating its success in accurately classifying arrhythmias.