Detecting structural heart disease from electrocardiograms using AI

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Early detection of structural heart disease is crucial for better patient outcomes.
• Current screening methods like echocardiography are limited by cost and accessibility.
• Previous AI models for heart disease detection were often trained on limited populations or specific conditions.

Purpose of the Study:

• To introduce EchoNext, a deep learning model designed for broad structural heart disease detection.
• To evaluate the diagnostic accuracy and generalizability of EchoNext.
• To assess the potential of AI in expanding large-scale heart disease screening.

Main Methods:

• Developed EchoNext, a deep learning model trained on over 1 million heart rhythm and imaging records.
• Validated the model's performance internally and externally.
• Conducted a prospective clinical trial on patients without prior cardiac imaging.
• Compared EchoNext's performance against cardiologists in a controlled setting.

Main Results:

• EchoNext demonstrated high diagnostic accuracy across diverse populations and care settings.
• The model outperformed cardiologists in a controlled evaluation.
• Prospective trial showed successful identification of previously undiagnosed heart disease.
• Consistent performance was observed across different racial and/or ethnic groups.

Conclusions:

• Deep learning models like EchoNext show significant potential for improving access to structural heart disease screening.
• AI can help overcome limitations of traditional imaging tools in widespread screening.
• Public release of model weights and data supports further research and transparency in AI for cardiovascular health.