Generation of Atrial Fibrillation Electrocardiogram and Atrial Fibrillation Classification Based on BiLSTM-CNN Generative Adversarial Network

In the medical field, data acquisition and sharing are facing severe challenges, which are mainly attributed to the high sensitivity of medical data and strict access restrictions. In order to resolve the issue, researchers have explored a variety of de-identity methods to promote data sharing, but these methods are sometimes inadequate in protecting personal privacy. Given this, the study in question focuses on using Generative Adversarial Networks (GAN) to generate realistic medical time series data, aiming to create a data solution that can protect privacy and meet the needs of research. The purpose of this study is to generate atrial fibrillation signals that accurately reflect the characteristics of normal electrocardiogram (ECG) waveforms. The generated synthetic time series data is expected to serve as an effective substitute for real patient data, thus facilitating its wide application in the medical field without compromising privacy. In order to achieve this goal, this paper uses a variety of GAN architectures to perform deep learning and generation of atrial fibrillation ECG. Subsequently, the authenticity and reliability of the generated data are quantitatively analyzed by three evaluation metrics to ensure its applicability in real-world applications. At the same time, integrating the generated synthetic data into the training set of the AF classifier has resulted in a significant improvement in the performance of the AF classifier. In summary, this paper not only demonstrates the potential of GAN for generating realistic medical time series data, but also the proposed method can be extended to simulate real ECG waveforms of other types of cardiovascular diseases.