Proof

Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence | Mayo Clinic

Written by Ultromics | Feb 8, 2024 1:13:13 PM

Ashley P. Akerman, Mihaela Porumb, Christopher G. Scott, Arian Beqiri, Agisilaos Chartsias, Alexander J. Ryu, William Hawkes, Geoffrey D. Huntley, Ayana Z. Arystan, Garvan C. Kane, Sorin V. Pislaru, Francisco Lopez-Jimenez, Alberto Gomez, Rizwan Sarwar, Jamie O'Driscoll, Paul Leeson, Ross Upton, Gary Woodward, and Patricia A. Pellikka.

Key findings

  • EchoGo® Heart Failure demonstrated strong diagnostic performance using only one routine echo video clip.
  • In independent testing, EchoGo® Heart Failure achieved 87.8% sensitivity and 81.9% specificity.
  • EchoGo® Heart Failure produced fewer nondiagnostic results than established clinical scores.
  • EchoGo® Heart Failure correctly reclassified almost three-quarters of patients whose HFA-PEFF or H2FPEF scores were indeterminate.
  • Patients classified as HFpEF by EchoGo® Heart Failure had significantly higher mortality, suggesting the model identified clinically meaningful disease risk.

Single echo video analysis

The study evaluated whether EchoGo® Heart Failure could detect HFpEF using only a single apical 4-chamber transthoracic echocardiogram video clip, a view routinely captured during standard echo exams.

This is important because HFpEF detection often requires multiple clinical and imaging inputs, which can be incomplete, discordant, or difficult to interpret. By using a routinely acquired video clip, EchoGo® Heart Failure may help reduce reliance on complex manual measurements and support more consistent identification of patients who need further evaluation.

Diagnostic performance

EchoGo® Heart Failure was trained and validated using data from Mayo Clinic and St George's University Hospitals NHS Foundation Trust, then tested in an independent multisite Mayo Clinic Health System dataset.

In independent testing of 1,284 patients, EchoGo® Heart Failure achieved 87.8% sensitivity and 81.9% specificity, with only 7.3% of studies classified as nondiagnostic because of high model uncertainty.

These findings demonstrate strong diagnostic performance using a single routinely acquired echocardiogram video and support the potential of EchoGo® Heart Failure to help identify HFpEF in clinical settings where diagnosis can be complex or uncertain.

Clinical score comparison

The study compared EchoGo® Heart Failure with established HFpEF clinical scoring systems, including HFA-PEFF and H2FPEF.

Both clinical scores produced high numbers of indeterminate results. EchoGo® Heart Failure correctly reclassified 73.5% of patients with indeterminate HFA-PEFF results and 73.6% of patients with indeterminate H2FPEF results.

This suggests EchoGo® Heart Failure may be particularly useful when conventional diagnostic pathways leave uncertainty.

Mortality findings

The study also found that patients classified as HFpEF by EchoGo® Heart Failure had higher age-adjusted mortality than those classified as not suggestive of HFpEF.

This supports the clinical relevance of the model’s output, showing that it was not only detecting imaging patterns, but also identifying patients with worse outcomes.

What this means for HFpEF diagnosis

HFpEF is a complex and heterogeneous condition that can be difficult to diagnose consistently, especially when traditional measurements are incomplete or indeterminate.

This study supports the potential of AI-enabled echocardiography to help automate detection, reduce diagnostic uncertainty, and support more timely identification of patients who may benefit from further assessment and treatment.

Publication

Akerman AP, Porumb M, Scott CG, et al. Automated Echocardiographic Detection of Heart Failure With Preserved Ejection Fraction Using Artificial Intelligence. JACC Advances. 2023;2(6):100452.