Evaluation demonstrates electrophysiological features captured by AI-ECG in cardiac assessment, indicating key structural correlates of biomarker estimation.
Key Points
To elucidate the specific electrocardiographic patterns and underlying cardiac abnormalities captured by artificial intelligence when predicting NT-proBNP levels from standard ECGs.
Assessed deep learning AI-ECG models developed to estimate circulating NT-proBNP biomarker levels.
Interpreted model focus and feature attribution across ECG waveforms to connect model predictions with cardiac morphology and hemodynamics.
AI-ECG models capture subtle voltage, repolarization, and conduction anomalies reflecting myocardial strain and elevated ventricular filling pressures.
Predicted NT-proBNP values correlate with subclinical structural remodeling that extends beyond conventional clinical ECG interpretation criteria.