AI could assist novice users obtain diagnostic-grade echocardiograms
Key takeaways:
- AI-assisted echocardiography could aid novice users in acquiring diagnostic-grade images.
- The AI tool demonstrated high concordance with images obtained by experts.
CHICAGO — AI could assist novice users of point-of-care ultrasound to acquire diagnostic-quality echocardiograms with high concordance vs. experts, according to new data presented at the American College of Cardiology Scientific Session.
“Echocardiography delivers dynamic imaging of the heart and provides essential data to guide clinical decisions in real time. A key limitation to the use of echocardiograms in many clinical settings is the lack of highly trained sonographers. In recent decades, with the rapid miniaturization and increased affordability of ultrasound systems, point-of-care ultrasound, or POCUS, performed in a variety of resource-constrained settings such as emergency departments, ICUs and medically underserved areas, has been gaining rapid popularity,” Caroline Ong, MD, MS, cardiologist at Northwell Health, said during a presentation. “However, cardiac POCUS performed by nonexpert users may be nonuniform in quality and run the risk of generating nondiagnostic and potentially misleading images, resulting in diagnostic and management errors. In light of this, several scientific teams have taken advantage of AI and advances in machine and deep learning to develop software that guides novices in their cardiac ultrasound exams.”
The software (HeartFocus, DESKi) features multiple algorithms designed for diagnostic-quality view detection, live guidance to assist in probe positioning and auto-recording of echocardiograms once positioning is deemed sufficient, Ong said.
Ong stated that these algorithms were trained on 1,483 patients and more than 1.2 million ultrasound images.
“HeartFocus is an AI-based software designed to guide novices with minimal upfront training to obtain diagnostic quality echocardiographic images. The live guidance feature here provides real-time cues to the users on how to move the probe on the patient’s chest to achieve the target positions required for each of the 10 echocardiographic reference views,” Ong said. “Once the AI software detects optimal positioning of the probe and the image reaches its diagnostic quality threshold for an adequate time, an automatic recording of the clip occurs. Additionally, the AI software continually assesses clip quality during scanning, and if a suitable clip for auto record cannot be captured, the software allows the users to retrospectively go back and record the highest quality clip obtained, referred to as the best effort record.”
For this prospective, noninferiority diagnostic imaging study conducted at Northwell Health and Centre Hospitalier Universitaire de Bordeaux in France, 240 patients were enrolled and presented for a clinically indicated echocardiogram (mean age, 63 years; 49% women; 39% with overweight; 21% with obesity). The performance goal for noninferiority was 80%, Ong said.
The primary objective was to assess whether the AI-based software could assist novice users to provide sufficient quality echocardiographic analyses of four parameters: Left ventricular size, LV function, right ventricular size and presence of nontrivial pericardial effusion.
At baseline, approximately 70% of participants had a known cardiac abnormality and 18% had implanted cardiac devices.
For the primary endpoint, AI-assisted echocardiography demonstrated a 100% performance goal compared with expert analysis for all four echocardiographic parameters.
The AI-assisted novices also achieved a performance goal of 95.4% or higher for parameters including RV function, left atrium size, segmental kinetics, aortic valve assessment, mitral valve assessment and tricuspid valve assessment.
The only parameter that fell below the performance goal for noninferiority was assessment of inferior vena cava size (78.3%) with AI-assisted novices compared with expert echocardiogram analysis.
Echocardiogram acquisition time was on average 23.6 minutes for the novices using AI-assisted analysis.
Ong said there were no adverse events or device deficiencies during the trial.“We demonstrate that AI-based software can enable novices to acquire diagnostic quality echocardiographic images for the accurate evaluation of important cardiac parameters. Novices could perform AI-guided exams with equal concordance to the expert sonographers and so met the standard for noninferiority for the primary endpoint,” Ong said during the presentation. “With the assistance of AI guidance, the learning curve for obtaining a diagnostic echocardiogram can be shortened compared with the current training done by sonographers and by cardiologists. In conjunction with a handheld portable ultrasound system, AI-based software can expand access of echocardiography to a variety of settings and times such as in rural medical practices or during patient home visits. However, while the study was carried out at two large teaching hospitals, further validation of use in different settings will be needed to reinforce the study's findings.”
Collapse
Ong C, et al. Clinical and investigative horizons I. Presented at: American College of Cardiology Scientific Session; March 29-31, 2025; Chicago (hybrid meeting).
Read more from American College of Cardiology
View all
Read more about