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

Background/Objectives: Vegetation length is a guideline-endorsed criterion for surgery in left-sided infective endocarditis (LSIE). However, its measurement is highly variable with crucial implications for decision making. A standardized measurement system would not only facilitate decision making in these patients, but also cutoff-point optimization for improving clinical guidelines. For this purpose, this work introduces an Artificial Intelligence (AI)-based system capable of extracting vegetation length from standard transesophageal echocardiography (TEE). Methods: Five echocardiographers independently measured the vegetation length of 76 vegetations from 67 consecutive patients with LSIE using offline TEE images. The AI-based system was trained on a multicenter registry with 353 patients with LSIE, comprising 282,096 echocardiographic frames annotated by an independent expert. This system was applied to measure vegetation length as an independent observer. The variability and correlation between operators and the system were assessed. Results: Lin's concordance correlation coefficient between the AI-based system and the mean of the measurements obtained by the echocardiographers was 0.74, comparable to the coefficient obtained between the echocardiographers themselves. Bland-Altman analysis showed a mean difference of -1.0 mm between the AI-based system and the mean of the measurements obtained by the echocardiographers. Conclusions: The AI-based system for vegetation measurement demonstrates a high level of correlation and agreement with experts, similar to the concordance between human operators themselves. This level of agreement suggests that the method has the potential to reduce inter-operator variability in the measurement process. Whether AI-based vegetation size improves embolism prediction has to be further investigated.