Cardiovascular diseases (CVDs) remain a leading cause of global morbidity and mortality. Standard 12-lead ECGs are widely accessible tools for cardiovascular screening and diagnosis; however, traditional visual interpretations have limitations. Recent advancements in deep learning offer unprecedented performance in extracting hidden patterns from ECG signals.
This study is a retrospective, multicenter investigation conducted across multiple tertiary medical centers in South Korea. The primary objectives are:
* To extract and encrypt raw 12-lead ECG data (XML format) into binary datasets to ensure patient privacy and data standardization.
* To advance and validate deep neural network models-specifically combining Bi-LSTM and attention mechanisms-for diagnosing conditions such as coronary artery disease, heart failure, and related structural heart diseases.
* To assess the clinical utility and prognostic performance of AI-ECG risk prediction models in real-world cohorts spanning data from 2005 to 2022.
