SGECG: A StarGAN-Based Framework for Intelligent ECG Generation and Augmentation

Electrocardiogram (ECG) serves as a critical bioelectrical signal for assessing cardiovascular system function, playing an indispensable role in the diagnosis, monitoring, and prognostic evaluation of cardiovascular diseases. However, traditional 12-lead ECG systems face several limitations, such as complex electrode configurations, inconvenient data collection, and the insufficiency of single-lead information for comprehensive patient assessment. To address these issues, this study proposes an innovative multi-lead ECG generation model based on Star-GAN, termed SGECG. This model employs a loss function capable of approximating target values to replace the discriminator, leveraging the architectural advantages of Star-GAN to achieve cross-lead ECG data transformation and enhance the efficiency and accuracy of data collection. We comprehensively evaluated the authenticity and reliability of the generated samples by assessing the quality of a pre-trained classifier on the generated multi-lead ECG data and measuring the performance of classifiers trained on synthetic data using cross-validation. Experimental results demonstrate that the SGECG model can synthesize ECGs that are highly consistent with real ECGs from other leads in terms of waveform features, time intervals, and amplitude, significantly enhancing the clinical applicability and diagnostic accuracy of ECG data.