Risk stratification system of gastrointestinal stromal tumors under EUS elastography based on artificial intelligence

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Gastrointestinal stromal tumors (GISTs) require preoperative risk stratification due to their malignant potential.
• Endoscopic ultrasound elastography (EUS-E) assesses tissue stiffness but current methods are subjective.
• An objective tool is needed for accurate GIST risk stratification using EUS-E.

Purpose of the Study:

• To develop and evaluate an AI-based system for objective GIST identification and risk stratification.
• To distinguish GISTs from other submucosal tumors (SMTs) using EUS-E.
• To assess the malignant potential of GISTs objectively.

Main Methods:

• Retrospective collection of 110 GIST cases and 189 SMT cases undergoing EUS-E.
• Construction of AI classification and segmentation models using 2625 EUS B-mode images.
• Extraction of an elasticity value (EUS-E-AI) from elastography images for risk assessment.

Main Results:

• AI classification and segmentation models achieved high accuracy (95.8%) and Dice coefficient (0.967).
• The EUS-E-AI value significantly differed between low-risk (0.268) and high-risk (0.186) GIST malignancy groups (P < 0.001).
• A cutoff EUS-E-AI of 0.224 accurately differentiated risk groups with 92.6% accuracy.

Conclusions:

• An AI-based system and elasticity indicator provide accurate and objective identification of GISTs.
• The developed system effectively stratifies GISTs based on malignant potential using EUS-E.
• This AI tool aids in clinical decision-making for GIST management.