Comprehensive echocardiogram evaluation with view primed vision language AI.

Area of Science:

• Cardiovascular imaging
• Artificial intelligence in medicine
• Medical diagnostics

Background:

• Echocardiography is a primary cardiac imaging tool.
• Current AI models in echocardiography are often limited to single views and tasks.
• This limits their performance and application scope.

Purpose of the Study:

• To develop a multi-view, view-informed AI foundation model for comprehensive echocardiogram interpretation.
• To address the limitations of single-view AI models in echocardiography.

Main Methods:

• Developed EchoPrime, a video-based vision-language foundation model trained on 12 million echocardiogram video-report pairs.
• Employed contrastive learning for a unified embedding model across all standard echocardiogram views.
• Integrated view classification and anatomical attention for holistic interpretation using retrieval-augmented methods.

Main Results:

• EchoPrime achieved state-of-the-art performance on 23 diverse cardiac form and function benchmarks across five international healthcare systems.
• Outperformed both task-specific AI approaches and prior foundation models.
• Demonstrated ability to integrate information from all echocardiogram videos for comprehensive interpretation.

Conclusions:

• EchoPrime represents a significant advancement in AI for echocardiography, moving beyond single-view limitations.
• The model shows potential to assist physicians in automated preliminary assessments of comprehensive echocardiograms.
• This technology can enhance the efficiency, reproducibility, and precision of cardiac diagnostics.