Multicenter validation study for automated left ventricular ejection fraction assessment using a handheld ultrasound...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate assessment of left ventricular ejection fraction (LVEF) is crucial for diagnosing and managing heart failure.
• Current standard methods for LVEF assessment often require experienced sonographers and high-end equipment.
• Point-of-care ultrasound (POCUS) with AI assistance offers potential for streamlined cardiac assessment.

Purpose of the Study:

• To validate a novel handheld ultrasound device with an artificial intelligence program (AI-POCUS) for automatic LVEF assessment.
• To compare AI-POCUS LVEF measurements against standard biplane disk methods.
• To evaluate the diagnostic performance of AI-POCUS in detecting reduced LVEF.

Main Methods:

• Prospective, multicenter study involving 200 patients across two Japanese hospitals.
• AI-POCUS was used for automatic LVEF assessment, compared with standard echocardiography.
• Intraclass correlation coefficient (ICC), bias, limits of agreement, sensitivity, and specificity were calculated.

Main Results:

• Analysis of 182 patients showed good correlation between AI-POCUS and standard LVEF (ICC = 0.81, p < 0.001) with minimal bias.
• AI-POCUS detected reduced LVEF (<50%) with 85% sensitivity and 81% specificity.
• AI-POCUS tended to underestimate left ventricular volumes, particularly for larger ventricles, though newer software versions improved this.

Conclusions:

• AI-POCUS demonstrates accurate LVEF assessment in a real-world setting, offering a promising tool for point-of-care cardiac evaluation.
• While LVEF assessment is reliable, careful consideration of LV volume measurements is advised.
• Continuous software improvement through large, diverse datasets enhances AI-POCUS performance, highlighting the value of big data in medical AI development.