Multi-center retrospective cohort study applying deep learning to electrocardiograms to identify left heart valvular...

Area of Science:

• Cardiology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Aortic Stenosis and Mitral Regurgitation are prevalent valvular heart conditions.
• Current diagnostic methods may not fully capture the disease burden.
• There is a need for advanced tools to aid clinical decision-making.

Purpose of the Study:

• To develop and validate deep learning (DL) models for screening and diagnosing Aortic Stenosis and Mitral Regurgitation.
• To leverage Electrocardiograms (ECGs) for valvular heart disease detection.
• To assess the clinical utility of DL tools in guiding patient management.

Main Methods:

• A multi-center retrospective cohort study utilizing Transthoracic Echocardiogram reports and ECGs from diverse patient populations.
• Development of a Natural Language Processing pipeline to extract ground-truth labels for valvular status.
• External validation of DL models for detecting Aortic Stenosis and Mitral Regurgitation.

Main Results:

• DL model for Mitral Regurgitation detection achieved an AUROC of 0.88 (internal) and 0.81 (external).
• DL model for Aortic Stenosis detection achieved an AUROC of 0.89 (internal) and 0.86 (external).
• Model performance improved leading up to diagnostic echocardiograms and showed promise for Transcatheter Aortic Valve Replacement (TAVR) evaluations.

Conclusions:

• Deep learning tools can extract valuable information from routine ECGs.
• These AI-driven tools are cost-effective and facilitate earlier disease detection.
• The application of DL in ECG analysis holds potential for improving patient prognosis in valvular heart disease.