Ultromics' ultrasound AI scores FDA clearance for spotting heart failure with preserved ejection fraction

Ultrasound artificial intelligence developer Ultromics scored an FDA green light for its program to detect hidden cases of heart failure.

The agency cleared the University of Oxford spinout’s EchoGo solution to help diagnose heart failure with preserved ejection fraction—where the cardiac muscle is strong enough to push a normal amount of blood out to the rest of the human body, but the patient is still at a high risk for hospitalization or death.

This preserved ejection fraction can make the underlying condition hard to spot using traditional measures, and Ultromics estimates that as many as 75% of cases can be misdiagnosed.

Developed in collaboration with the Mayo Clinic, the program uses AI to analyze images from an echocardiogram for the more subtle signs of muscle strain or higher blood pressure within the heart’s chambers. 

Heart failure with preserved ejection fraction, or HFpEF, is typically found by a specialized analysis of the heart’s diastolic function—examining the flow of blood within the lull between two heartbeats—which can sometimes be inaccurate.

The clearance of a simpler, AI-powered HFpEF diagnostic is especially well-timed, following the recent approvals of new drug therapies for the condition, Ultromics said in a press release. Eli Lilly and Boehringer Ingelheim’s SGLT2 inhibitor Jardiance received expanded heart failure approvals in the U.S. and Europe earlier this year.

The FDA’s green light for EchoGo, which previously received a breakthrough designation from the agency, also comes shortly after Ultromics’ entry into the Accelerating Medicines Partnership Heart Failure program led by the Foundation for the National Institutes of Health. AMP HF is a five-year, public-private research collaboration aimed at HFpEF, with the goal of finding more specific subtypes of the condition and potentially targeted therapies.

According to Ultromics, EchoGo demonstrated 90% accuracy in detecting HFpEF within a validation dataset, with a false-negative rate of 12.2% and a false-positive rate of 17%. The company said its AI was able to correctly identify 68% more HFpEF patients compared to traditional clinical algorithms in an independent testing dataset.