Modular eFAST tissue phantom for AI-based ultrasound triage

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Ultrasound (US) imaging is crucial in emergency medicine and battlefield triage due to its portability.
• Interpreting and acquiring US images requires specialized training, presenting a barrier to widespread use.
• Artificial intelligence (AI) integration can enhance US imaging interpretation and diagnostic accuracy.

Purpose of the Study:

• To develop a full torso tissue-mimicking phantom for simulating extended-focused assessment with sonography for trauma (eFAST) examinations.
• To create a platform for developing AI guidance and classification models for US imaging.
• To improve AI model performance in detecting anatomical features and diagnosing injury states using simulated US data.

Main Methods:

• A full torso tissue-mimicking phantom was engineered to simulate eFAST scan sites and thoracic motion.
• The phantom was used to capture US images, which were then utilized to train AI models.
• AI models were developed for anatomical feature detection and injury state diagnosis based on phantom-acquired images.

Main Results:

• The phantom successfully simulated thoracic motion and modular injuries across all eFAST scan sites.
• AI models trained on phantom data achieved Intersection over Union (IOU) scores greater than 0.80.
• The AI models demonstrated a 71.5% accuracy in diagnosing injury states on blind inference tests.

Conclusions:

• The developed tissue-mimicking phantom is a reliable tool for generating eFAST images for AI model training.
• The phantom can be utilized for training medical personnel in US examination techniques.
• This technology supports the development of automated US image acquisition techniques.