A new AI approach could help clinicians identify patients with advanced heart failure more efficiently, using data already available in routine care. The findings come from a collaboration between researchers at Weill Cornell Medicine, Cornell Tech, Columbia University Vagelos College of Physicians and Surgeons and NewYork-Presbyterian.
Advanced heart failure is typically diagnosed using cardiopulmonary exercise testing (CPET), a resource-intensive procedure that requires specialised equipment and expertise and is often limited to large medical centres. As a result, many patients remain undiagnosed or do not receive appropriate care.
Addressing limitations
The newly developed AI model aims to address this limitation by predicting peak oxygen consumption (peak VO2), a key measure used in CPET, based on cardiac ultrasound images combined with data from electronic health records.
“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 author Dr. Fei Wang of Weill Cornell Medicine. The findings were recently published in npj Digital Medicine.
Multimodal AI model
The research is part of the Cardiovascular AI Initiative, a joint effort by Cornell, Columbia and NewYork-Presbyterian to explore the role of AI in heart failure care. The team developed a multimodal machine learning model capable of analysing multiple data types, including ultrasound video of the heart, waveform data related to blood flow and valve function, and clinical information from patient records. The model was trained on anonymised data from 1,000 heart failure patients and subsequently tested on an independent group of 127 patients across multiple hospital sites.
In validation, the AI model achieved an accuracy of approximately 85 percent in identifying high-risk patients based on predicted peak VO2 values. According to the researchers, this level of performance suggests the model could be clinically useful for screening and triage.
Importantly, the system may enable earlier identification of patients who would otherwise not be recognised due to limited access to specialised diagnostic testing.
Close collaboration
The researchers emphasise that the project highlights the importance of close collaboration between clinicians and AI experts. “This was a case of medicine shaping the future of AI, not just AI shaping the future of medicine,” said Dr. Deborah Estrin of Cornell Tech.
Further clinical studies are planned to validate the approach and support regulatory approval. If successful, the technology could help improve access to advanced heart failure diagnostics and ultimately lead to better patient outcomes.
“If we can use this approach to identify many advanced heart failure patients who would not be identified otherwise, then this will change our clinical practice and significantly improve patient outcomes and quality of life,” said Dr. Nir Uriel of NewYork-Presbyterian.
AI ultrasound
In 2024, Philips introduced an AI-powered platform for cardiovascular ultrasound that aims to improve diagnostic accuracy while reducing workload for clinicians. The solution, integrated into the EPIQ CVx and Affiniti CVx systems and cleared by the US FDA, automates key analysis steps and accelerates workflows. This enables faster and more consistent interpretation of cardiac images, potentially reducing the need for repeat scans.
The technology supports earlier detection of conditions such as heart failure, which affects millions worldwide. AI algorithms can also accurately identify regional wall motion abnormalities, an important indicator of cardiovascular disease. By automating measurements and image interpretation, the platform improves efficiency and reproducibility, allowing clinicians, regardless of experience, to analyse ultrasound data more quickly and reliably, ultimately supporting better patient outcomes.
