AI-Driven Cardiovascular Imaging Research Showcased at ASE 2023

During the American Society of Echocardiography’s (ASE) 34th Annual Scientific Sessions, researchers highlighted how artificial intelligence (AI) may enhance cardiovascular imaging by helping clinical staff acquire diagnostic-quality images, flagging heart failure with preserved ejection fraction (HFpEF), and determining aortic stenosis severity.

The first study, titled ‘Real-Time Artificial Intelligence Based Guidance of Echocardiographic Imaging: Does Novice Profile Affect Image Quality and Suitability for Diagnostic Interpretation?’ examined whether AI could help novice nurses and medical residents acquire diagnostic-quality echocardiography images.

In the study, researchers compared the quality of echocardiography images acquired by nurses and medical residents with and without AI assistance. The model was developed using visual cues from over 10,000 echocardiograms and embedded into a handheld device for users.

The research team found that after 12 hours of training with the AI tool, these users could acquire images in ten standard views that approached diagnostic quality rates similar to those of expert sonographers.

The accuracy of the AI was assessed by five blinded expert readers to determine whether the outputs achieved sufficient quality for diagnostic interpretation.

“We found that the AI-guided software allowed novices to acquire images suitable for diagnostic interpretation by an expert reader in the majority of patients,” explained Roberto M. Lang, MD, a lead author on the study and specialist in echocardiography at the University of Chicago Medicine, in the press release. “These promising findings confirm that AI-guided software could be potentially useful in teaching novice medical professionals and assessing cardiac function in settings where services of trained echocardiography laboratories are not available.”

The second study, ‘Automated Echocardiographic Detection of Heart Failure with Preserved Ejection Fraction using Artificial Intelligence,’ aims to address challenges associated with diagnosing HFpEF. The press release indicates that HFpEF creates significant disease burden worldwide, and Mayo Clinic reports that the prevalence of the condition is increasing in developed countries.

Algorithms to diagnose HFpEF using echocardiography measurements and clinical risk factors have been developed, but these are indeterminate for many patients. To address this, the researchers created a deep learning (DL) model to analyze single, routinely acquired echocardiographic video clips to help better identify patients with the condition.

“Our novel AI model demonstrated excellent discrimination of patients with HFpEF versus without, more often than with clinical scores, and it was able to stratify patients according to mortality risk,” said senior author Patricia A. Pellikka, MD, chair of the Division of Cardiovascular Ultrasound at Mayo Clinic in Rochester, Minn. “Further testing in varied populations and in different echocardiography labs is needed to refine the model, but it could potentially be used for screening and importantly, expedite diagnosis and treatment for a large group of patients.”

The third study, titled ‘Fully Automated Artificial Intelligence Assessment of Aortic Stenosis by Echocardiography,’ evaluated an AI tool’s ability to measure echocardiographic markers of aortic stenosis severity without the need for human input.

Aortic stenosis is a serious and common heart valve disease that occurs when the aortic valve opening narrows and restricts blood flow from the left ventricle to the aorta, according to the American Heart Association. The condition is estimated to be present in over 12 percent of older Americans.

Determining the severity of aortic stenosis is crucial to improve patient outcomes, but doing so necessitates the use of transthoracic echocardiography, a cardiovascular imaging method that requires sonographic expertise to accurately interpret.

AI tools to overcome this hurdle have been developed, the press release states, but none have been used to assess aortic stenosis without human input.

The AI used in the study leveraged this ‘hands off’ approach, and its outputs were highly accurate compared to expert cardiologist assessment.

These findings suggest that similar tools could help with the rapid identification of hemodynamically significant aortic stenosis, reducing the time to care while preventing unnecessary morbidity and mortality, the researchers noted.

“AI technology can allow for near instantaneous and hands-free identification of a life-threatening valvular heart condition by providers trained only in basic ultrasound,” said lead author Hema Krishna, MD, a cardiologist at the University of Illinois at Chicago. “This machine learning platform can potentially extend diagnostic capability to patients in rural, community, or emergency room settings without immediate access to cardiologists.”

Other research presented at this year’s ASE conference includes studies examining how AI-powered chatbots can help improve patient understanding of echocardiography reports, exploring how advances in echocardiography can lead to more accurately placed pacemakers, and demonstrating how echocardiography can be used to analyze left ventricular global longitudinal strain (GLS) among college athletes with COVID-19.