31 July 2023. A system using deep learning algorithms to analyze ultrasound video images is shown to detect a difficult-to diagnose type of heart failure with clips from one view of the heart. Researchers at the Mayo Clinic in Rochester, Minnesota and heart disease analytics company Ultromics Ltd. in Oxford, U.K. describe their findings in today’s issue of the journal JACC: Advances, and presented last month at a meeting of the American Society of Echocardiography.
Heart failure is a common and sometimes deadly condition, where the heart cannot pump enough blood to the rest of the body to support other organs. Centers for Disease Control and Prevention estimates some 6.2 million people in the U.S. have heart failure, with nearly 380,000 deaths attributed to the disease in 2018. People with coronary artery disease, high blood pressure, diabetes, and obesity are considered at the highest risk of heart failure.
One indicator of heart failure is reduced blood flow from the heart’s left ventricle, known as ejection fraction. The left ventricle pumps blood through the aorta providing oxygen for cells, tissue, and organs in the body. In some cases, the heart appears to be pumping normally, but the amount of blood pumped from the left ventricle is reduced, known as heart failure with preserved ejection fraction, or HFpEF. Because of its complexity, say the journal paper’s authors, HFpEF can be difficult to detect with conventional echocardiograms, an ultrasound technique for detecting heart disease in real time.
Ultromics is a six year-old company with a technology based on research at the University of Oxford lab of Paul Leeson, professor of cardiovascular medicine, who studies early indicators of heart disease. Among the lab’s work is applying artificial intelligence to improve heart disease diagnostics with echocardiograms. Ultromics collaborates with the Mayo Clinic on EchoGo Heart Failure, the company’s lead system for diagnosing HFpEF from echocardiogram images.
Combining image analysis and deep learning
Leeson co-founded Ultromics with former doctoral student Ross Upton in 2017, and serves as the company’s chief medical officer, while Upton is CEO. In Dec. 2022, Science & Enterprise reported on the EchoGo system gaining FDA authorization for marketing in the U.S.
In their paper, a team led by Mayo Clinic cardiologist Patricia Pellikkia, also director of the clinic’s echocardiography lab, describes development of the EchoGo model. The authors say EchoGo is based on algorithms in a convolutional neural network, a type of artificial intelligence that combines image analysis and deep learning. The algorithms discern components in images for training deep-learning analytics to reveal underlying patterns and relationships in those image components. As the algorithms encounter more images and components, the algorithms become better able to analyze new images.
In this case, EchoGo is trained with echocardiogram clips viewed from the apex or top of the heart chamber, capturing real-time video of all four heart chambers, called a transthoracic echocardiograph or TTE. The authors report training the model with TTE video from 2,971 individuals at the Mayo Clinic with HFpEF and 3,785 without the condition. In an independent validation with 1,284 individuals, divided about evenly between patients with and without HFpEF, EchoGo returned a true-positive sensitivity of 88 percent and true-negative specificity of 82 percent. Applying the model to more than 700 cases where conventional heart failure tests previously returned indeterminate results, EchoGo correctly identified 74 percent of individuals with HFpEF.
“HFpEF can be difficult to detect,” says Pellikkia in an Ultromics statement, “but left undetected and untreated, can result in hospitalization and mortality. As the first A.I. platform cleared to detect the condition, EchoGo Heart Failure can fill a significant unmet need.”
More from Science & Enterprise:
- Small Biz Award to Boost Maternal Health App
- Digital Biomarkers Shown to Highlight Parkinson’s Disease
- Automated Life Science A.I. Generation System Unveiled
- Neuro Biomarkers, Home EEG Assessed in Upcoming Trial
- Rare Disease Search Adds Large Language Model A.I.
We designed Science & Enterprise for busy readers including investors, researchers, entrepreneurs, and students. Except for a narrow cookies and privacy strip for first-time visitors, we have no pop-ups blocking the entire page, nor distracting animated GIF graphics. If you want to subscribe for daily email alerts, you can do that here, or find the link in the upper left-hand corner of the desktop page. The site is free, with no paywall. But, of course, donations are gratefully accepted.
[wpedon id=”42724″ align=”center”]
* * *
You must be logged in to post a comment.