Artificial intelligence-based automated interpretation of images of electrocardiograms: development and multinational validation of ECG-GPT - PubMed
7 hours ago
- #Electrocardiography
- #Digital Health
- #Artificial Intelligence
- Development of ECG-GPT, an AI model for expert-level interpretation of ECG images.
- Validation across multinational datasets with 4.1 million ECGs, showing high diagnostic accuracy (AUROCs 0.80-0.96).
- ECG-GPT achieves a median pairwise similarity of 0.90 for diagnosis statements, outperforming baselines.
- Model performance consistent across diverse health settings, including US, Brazil, UK, and Germany.
- Potential to address ECG interpretation bottlenecks in low-resource settings.
- Authors disclose multiple patents, funding sources, and conflicts of interest related to AI and healthcare technologies.