A Deep Learning Approach for Patient-Specific ECG Signal Classification with 1-D CNNs

Abstract

Cardiovascular diseases (CVDs) still symbolize the leading cause of death in the world. Electrocardiogram (ECG) is a primary diagnostic modality for diagnosing conditions, but its interpretation requires clinical expertise and is prone to inter-observer variability. Recent advances in deep learning have enabled the development of automated ECG classification algorithms and, thus, have offered more stable and effective decision-support mechanisms. This paper presents a small one-dimensional Convolutional Neural Network (1D-CNN) used to detect four types of rhythm in patient-specific ECG signals: Normal, Atrial Fibrillation, Other, and Noise. The network uses very little preprocessing of the raw ECG recordings, making it an end-to-end learning pipeline. Using the PhysioNet/Computing in Cardiology Challenge 2017 dataset, three CNN configurations were evaluated. The proposed framework achieved an overall accuracy of 85.4%, a macro-F1 score of 0.81, and a weighted F1 Score of 0.803. Results demonstrate that a lightweight CNN can achieve competitive performance compared to more complex state-of-the-art methods, while maintaining simplicity, reproducibility, and potential for clinical deployment.