Standardizing Vegetation Size Measurement in Native Left-Sided Infective Endocarditis Using Artificial Intelligence

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Vegetation length is a key surgical criterion for left-sided infective endocarditis (LSIE).
• Current measurement methods are highly variable, impacting clinical decision-making and guideline adherence.
• Standardization is needed to optimize decision-making and improve clinical guidelines for LSIE.

Purpose of the Study:

• To introduce an Artificial Intelligence (AI)-based system for extracting vegetation length from transesophageal echocardiography (TEE).
• To assess the AI system's accuracy and consistency in measuring vegetation length compared to expert echocardiographers.
• To evaluate the potential of AI in reducing inter-operator variability in LSIE vegetation measurement.

Main Methods:

• An AI system was trained on a multicenter registry of 353 LSIE patients (282,096 frames).
• Five echocardiographers independently measured vegetation length in 76 vegetations from 67 LSIE patients.
• The AI system measured vegetation length on the same TEE images, acting as an independent observer.

Main Results:

• The AI system demonstrated a Lin's concordance correlation coefficient of 0.74 with the mean of expert measurements.
• This correlation is comparable to the concordance observed between human echocardiographers.
• Bland-Altman analysis revealed a mean difference of -1.0 mm between AI and expert measurements.

Conclusions:

• The AI-based vegetation measurement system shows high correlation and agreement with expert clinicians.
• The AI system's performance is similar to inter-observer agreement among echocardiographers.
• This AI tool has the potential to reduce measurement variability, though its impact on embolism prediction requires further study.