Researchers have developed a new artificial intelligence (AI) approach that may help doctors identify patients with advanced heart failure more easily and efficiently. The study, led by investigators from Weill Cornell Medicine, Cornell Tech, Cornell Ann S. Bowers College of Computing and Information Science, Columbia University Vagelos College of Physicians and Surgeons and NewYork-Presbyterian, suggests that AI could improve access to care for many patients whose condition may otherwise go undetected.
Advanced heart failure occurs when the heart can no longer pump blood effectively enough to meet the body’s needs. Diagnosing this condition often requires cardiopulmonary exercise testing (CPET), a specialised procedure that measures how well the heart and lungs function during physical activity. However, CPET requires expensive equipment and specially trained staff and is usually only available at large medical centres. As a result, many patients with advanced heart failure are not identified early enough to receive appropriate treatment. Researchers estimate that about 200,000 people in the United States are living with advanced heart failure, yet only a small proportion receive the specialised care they need each year.
In the new study, published in the journal npj Digital Medicine, researchers tested whether AI could help overcome this challenge. Instead of relying on CPET, the new method uses routine cardiac ultrasound images, also known as echocardiograms, along with information already contained in patients’ electronic health records. The AI system was designed to predict peak oxygen consumption, or peak VO2, which is considered one of the most important measurements obtained during CPET and a key indicator of advanced heart failure severity.
“This opens up a promising pathway for more efficient assessment of patients with advanced heart failure using data sources that are already embedded in routine care,” said senior study author Fei Wang, associate dean for AI and data science at Weill Cornell Medicine. The project was part of the Cardiovascular AI Initiative, a collaborative effort involving researchers and clinicians from Cornell, Columbia and NewYork-Presbyterian aimed at exploring how AI can improve the diagnosis and management of heart failure.
The researchers emphasised that the project relied on close collaboration between physicians and AI specialists. Nir Uriel, director of advanced heart failure and cardiac transplantation at NewYork-Presbyterian, explained that heart failure specialists first identified areas where AI could have the greatest clinical impact. One of the most promising opportunities was using AI to analyse cardiac ultrasound data to identify patients at high risk. Deborah Estrin, associate dean for impact at Cornell Tech, noted that the collaboration also helped inspire the development of new AI techniques driven directly by clinical needs.
To create the model, the research team trained the AI system using deidentified medical data from 1,000 patients with heart failure treated at NewYork-Presbyterian/Columbia University Irving Medical Center. The AI analysed several different types of information simultaneously, including moving ultrasound images of the heart, images showing blood flow and valve function, and data from electronic health records. After training, the model was tested on a separate group of 127 heart failure patients from three additional NewYork-Presbyterian campuses. The AI demonstrated an overall accuracy of about 85% in identifying patients at high risk, making it one of the strongest reported results so far for AI-based prediction of peak VO2.
The research team is now planning clinical studies to evaluate the technology further before it can be considered for approval by the U.S. Food and Drug Administration and eventual use in routine patient care. Researchers believe the approach could help doctors identify many patients with advanced heart failure earlier and more efficiently than current methods allow. According to Dr. Uriel, improving identification of these patients could ultimately lead to major improvements in treatment, quality of life and long-term health outcomes.
More information: Zhe Huang et al, Multimodal multi-instance learning for cardiopulmonary exercise testing performance prediction, npj Digital Medicine. DOI: 10.1038/s41746-026-02493-w
Journal information: npj Digital Medicine Provided by Weill Cornell Medicine
