Artificial intelligence-assisted echocardiography can help novices screen for heart failure
Presenter: Feiqiong Huang, MD, PhD, National Heart Centre Singapore, Singapore
Point-of-care AI-assisted novice echocardiography for screening of heart failure (PANES_HF). Abstract 1081-05.
Artificial intelligence (AI)-assisted point-of-care echocardiography enables laypeople to perform accurate echocardiography screening for heart failure in most cases, according to this study by researchers at the National Heart Centre in Singapore and Duke University, Durham, North Carolina.
They found that with 2-weeks of training, novice laypeople can accurately acquire 2-dimensional echocardiography images to screen for heart failure. By combining artificial intelligence (AI)-assisted echocardiography conducted by the novices and AI-automated image analysis, the AI-novice pathway provided results that were superior to tests measuring N-terminal-pro B-type natriuretic peptide (NT-proBNP) levels in detecting left ventricular ejection fraction (LVEF) less than 50%.
“The AI-novice pathway is feasible and effective for heart failure screening,” said presenter Feiqiong Huang, MD, PhD, of the National Heart Centre in Singapore.
The increasing prevalence of heart failure in aging populations is increasing the demand for echocardiography. However, there is a worldwide shortage of trained sonographers and long waiting times for expert echocardiograms.
In this study, researchers prospectively recruited 100 symptomatic patients with suspected heart failure. Laypeople with no prior echocardiography experience underwent 2 weeks of training to acquire echocardiography images with AI guidance using the EchoNous Kosmos handheld echocardiography system, with AI-automated image analysis by Us2.AI (AI pathway).
All patients also had standard cart-based echocardiography by trained sonographers and interpreted by qualified cardiologists, which served as a reference standard, and blood sampling for NT-proBNP. The primary endpoint of the study was the accuracy of the AI pathway in detecting abnormal scans, defined as LVEF less than 50%.
AI-assisted echocardiography yielded interpretable results in 96 patients. The area under the curve of the AI pathway was 0.88 (95% CI 0.802, 0.958), which is superior to the prior established NT-proBNP cutoffs for heart failure, said Huang. The integrated discriminant improvement of the AI pathway was 29.7%. Adding NT-proBNP to the AI pathway did not significantly improve the diagnostic accuracy.
There was good agreement between clinician and AI pathway detection of LVEF less than 50%, said Huang. The difference in LVEF between both arms was 2.3%. The median time required for 1 novice echocardiogram was 11 minutes and 28 seconds.
In conclusion, Huang said: “This study affirms the role of AI to empower a novice to perform point-of-care ultrasound to obtain accurate diagnostic information to detect heart failure. This augments the supply of echocardiogram services by minimizing the need for sonographic expertise in the traditional cart-based setting.”
Future studies involving a larger population and integration of this AI-enhanced novice pathway into primary care pathways are currently in progress.
“The AI-novice pathway has the potential to improve heart failure care,” Huang said. “As we begin to view heart failure across a spectrum of ejection fraction, with individualized therapies for each ejection fraction category, LVEF becomes an important marker in addition to NT-proBNP. The AI-novice pathway can increase accessibility to LVEF information, enabling early therapy initiation, which improves heart failure outcomes and circumvents long wait times for traditional cart-based echocardiography.”
References
Barry T, Farina JM, Chao CJ, et al. The role of artificial intelligence in echocardiography. J Imaging 2023; 9(2):50. doi:10.3390/jimaging9020050
Huang W, Lee A, Tromp J, et al. Point-of-care AI-assisted novice echocardiography for screening of heart failure (PANES-HF). J Am Coll Cardiol 2023; 81(8 Suppl):2145. https://doi.org/10.1016/S0735-1097(23)02589
Disclosures
Feiqiong Huang: Nothing to disclose