AI Guides Nonexperts to Expert-Level Lung Ultrasound

TOPLINE:

Artificial intelligence (AI)–assisted lung ultrasound enabled nonexperts to capture diagnostic-quality images after brief software-focused training, matching expert standards in vast majority of cases.

METHODOLOGY:

• Researchers conducted a prospective diagnostic-validation study between July and December 2023 to evaluate the ability of AI to guide healthcare professionals (HCPs) inexperienced in lung ultrasound acquire diagnostic-quality images.
• They included 176 participants (mean age, 63 years; 46% women) with concerns of pulmonary edema who underwent two lung ultrasound examinations — one performed by a trained HCP without formal lung ultrasound training who used Lung Guidance AI software and the other performed by a trained expert who did not use the AI.
• A panel of five expert lung ultrasound readers, blinded to the study operators and the use of AI, independently reviewed the image clips.
• The performance of AI was considered successful if at least 80% of the lung ultrasound clips met the clinical assessment standards.

TAKEAWAY:

• Overall, 98.3% (95% CI, 95.1%-99.4%) of ultrasound studies conducted by trained HCPs with the use of AI were considered of diagnostic quality by the expert panel.
• No statistically significant differences were observed in the overall study-level image quality between studies conducted by trained HCPs with the use of AI and those conducted by lung ultrasound experts without the use of AI.
• For zone 6 (left antero-inferior region) specifically, trained HCPs outperformed lung ultrasound experts, acquiring 90.9% diagnostic-quality images compared with 77.3% by lung ultrasound experts (P < .001).

IN PRACTICE:

“By enabling a wide spectrum of healthcare professionals to conduct LUS [lung ultrasound] examinations, this AI solution has the potential to significantly enhance diagnostic accuracy and patient care,” the authors wrote. 

SOURCE:

This study was led by Cristiana Baloescu, MD, MPH, Yale University School of Medicine, New Haven, Connecticut. It was published online on January 15, 2024, in JAMA Cardiology. 

LIMITATIONS:

The quality of images acquired by trained HCPs with the help of AI were not directly compared with those obtained without the help of AI. Moreover, including ultrasound-trained physicians in the trained HCP group may have introduced bias, potentially favoring the overall results in favor of AI.

DISCLOSURES:

This study was funded by Caption Health (now GE HealthCare) through a grant from the Bill and Melinda Gates Foundation. Some authors reported receiving grants, consulting fees, and personal fees from Caption Health and various other pharmaceutical and healthcare companies.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.