Automatic Detection of Vegetations With Transesophageal Echocardiography in Infective Endocarditis Using Artificial...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Infective endocarditis (IE) diagnosis relies on detecting vegetations, often visualized via transesophageal echocardiography (TEE).
• Current manual analysis of TEE images for vegetations is limited by basic measurements and significant inter-operator variability.
• There is a critical need for objective and reproducible methods to improve IE diagnosis.

Purpose of the Study:

• To develop and evaluate an AI-based model for automated vegetation detection and diagnosis in TEE images.
• To compare the performance of different AI architectures (YOLO and DETR) for this task.
• To assess the potential of AI to assist non-expert cardiologists in IE diagnosis.

Main Methods:

• A retrospective observational study involving 329 IE patients across 7 hospitals.
• Training an AI model using YOLO and DETR architectures on TEE images to detect vegetations.
• Evaluating the AI model's performance at both the frame level (detection) and patient level (diagnosis).

Main Results:

• The AI model demonstrated strong diagnostic capability with an AUROC of 0.91 (PPV=0.81, TPR=0.83).
• Frame-level vegetation detection achieved promising results (PPV=0.83, TPR=0.75).
• Both YOLO and DETR architectures showed high performance in detecting vegetations and diagnosing IE.

Conclusions:

• The developed AI algorithm effectively detects vegetations and identifies patients with IE from TEE images.
• This AI tool has the potential to facilitate and accelerate IE diagnosis, particularly for non-expert cardiologists.
• AI-powered analysis offers improved accuracy and reduced variability compared to traditional methods.