AI-Powered Cardiac Ultrasound Analysis Successfully Predicts Advanced Heart Failure Risk Using Routine Clinical Data

New AI research from Weill Cornell and Columbia identifies advanced heart failure with 85% accuracy using standard ultrasounds and health records.

By: AXL Media

Published: Mar 24, 2026, 5:11 AM EDT

Source: Information for this report was sourced from Weill Cornell Medicine

Addressing the Diagnostic Bottleneck in Cardiac Care

Advanced heart failure affects an estimated 200,000 people in the United States, yet only a fraction receive appropriate specialized care each year. This disparity is largely due to the diagnostic requirement for cardiopulmonary exercise testing (CPET), a process that demands specialized equipment and highly trained staff typically found only at major medical centers. Researchers from Weill Cornell Medicine and Columbia University have now introduced a promising alternative: an artificial intelligence model that utilizes data already embedded in routine clinical practice. By leveraging common cardiac ultrasound data, the AI can identify patients at the highest risk, offering a pathway to better outcomes for thousands who might otherwise be overlooked.

Medicine Shaping the Future of Artificial Intelligence

The development of this technology was the result of the Cardiovascular AI Initiative, a multidisciplinary effort involving more than 40 heart failure specialists. Unlike many AI applications that are developed in isolation from clinical needs, this model was shaped by direct input from cardiologists who identified a critical gap in diagnostic accessibility. Dr. Deborah Estrin of Cornell Tech noted that this collaboration represents a case of medicine shaping the future of AI technology. The resulting machine learning model was developed over several years to ensure it could handle the complexities of real-world medical data, moving beyond simple image recognition to integrated clinical analysis.

A Multi-Modal Approach to Patient Risk Assessment

The AI team, led by Dr. Fei Wang, developed a multi-modal and multi-instance model capable of processing three distinct streams of information simultaneously. The system analyzes ordinary moving ultrasound images of the heart, waveform imagery depicting blood flow and valve dynamics, and relevant data points extracted from electronic health records. By synthesized these varied inputs, the AI can predict peak oxygen consumption (peak VO2), which is currently considered the most important measure for determining heart failure severity. This integrated approach allows the model to capture subtle patterns that might be missed by human observation or simpler diagnostic tools.