GenECG: A synthetic image-based ECG dataset to augment artificial intelligence-enhanced algorithm development

Abstract Background Artificial intelligence-enhanced electrocardiogram (AI-ECG) algorithms have primarily been developed using digitised signal data, due to a relative absence of image-based datasets. An image-based ECG dataset incorporating artefacts common to paper-based ECGs, which are typically scanned or photographed into electronic health records, could facilitate clinically useful AI-ECG algorithm development. This study aimed to create a high-fidelity, synthetic image-based ECG dataset to enable image-based AI-ECG algorithm development. Methods ECG images were re-created from the PTB-XL database, a publicly available signal-based dataset, and image manipulation techniques were applied to mimic artefacts associated with ECGs in real-world settings. To evaluate the fidelity of the synthetic images, iterative clinical Turing tests were conducted. The ability of currently available algorithms to analyse synthetic ECG images containing artefacts was assessed. Results Synthetic images were created from all PTB-XL ECGs leading to the creation of GenECG, an image-based dataset containing 21,799 ECGs with artefacts encountered in routine clinical care paired with artefact-free images. Clinical Turing tests confirmed the realism of the images: expert observer accuracy of discrimination between real-world and synthetic ECGs fell from 63.9% (95% CI 58.0%-69.8%) to 53.3% (95% CI: 48.6%-58.1%) over three rounds of testing, indicating that observers could not distinguish between synthetic and real ECGs. The performance of pre-existing image-based algorithms on synthetic (AUC 0.592, 95% CI 0.421-0.763) and real-world (AUC 0.647, 95% CI 0.520-0.774) ECG images containing artefact was limited. Algorithm fine-tuning with GenECG data led to an improvement in classification accuracy on real-world ECG images (AUC 0.821, 95% CI 0.730-0.913) demonstrating the potential for synthetic data to augment image-based AI-ECG algorithm development. Conclusions GenECG is the first synthetic image-based ECG dataset to pass a clinical Turing test. The dataset will enable image-based AI-ECG algorithm development, ensuring utility in low resource areas, pre-hospital settings and hospital environments where signal data are unavailable. What is already known on the subject? Artificial intelligence-enhanced ECG (AI-ECG) analysis presents a significant opportunity to improve the care of patients with cardiovascular disease. Most AI-ECG algorithms have been developed using ECG signal data, limiting their ability to analyse paper-based ECGs which are still prevalent in various hospital and non-hospital settings. What this study adds This study presents GenECG, a high-fidelity, synthetic dataset comprising 21,799 ECG images paired with artefact-free images and ECG signal data. GenECG is the first publicly available synthetic, image-based ECG dataset to pass a clinical Turing test. The performance of image-based AI-ECG algorithms improved through fine-tuning with GenECG data demonstrating the potential for synthetic data to augment AI-ECG research. How this study might affect research, practice or policy GenECG will facilitate the development of image-based AI-ECG algorithms, promising to expand the application of AI-ECG to traditional hospital settings, reliant on paper-based ECGs, and non-hospital environments such as remote healthcare areas or pre-hospital settings.