Using artificial intelligence to predict choledocholithiasis: can machine learning models abate the use of MRCP in...

Area of Science:

• Hepatobiliary Medicine
• Medical Informatics
• Artificial Intelligence in Healthcare

Background:

• Prompt diagnosis of choledocholithiasis is critical for patient outcomes and resource management.
• Magnetic resonance cholangiopancreatography (MRCP) is a common but costly diagnostic tool that can delay treatment.
• Developing predictive models can optimize patient care pathways for suspected choledochocholithiasis.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting choledocholithiasis risk.
• To compare ML model performance against existing clinical guidelines (ASGE).
• To identify clinical and imaging factors that predict choledocholithiasis.

Main Methods:

• Retrospective analysis of 222 inpatients with suspected choledocholithiasis.
• Utilized logistic regression, XGBoost, random forest, and K-nearest neighbours ML techniques.
• Developed an ensemble model from the top three performing ML models.

Main Results:

• Machine learning models demonstrated high accuracy (up to 0.81 AUROC) in predicting choledocholithiasis.
• All developed models outperformed the ASGE risk stratification guidelines.
• Key predictive factors included common bile duct diameter, lipase levels, cholelithiasis on imaging, and liver function tests.

Conclusions:

• ML models offer a viable tool for accurately assessing choledocholithiasis risk.
• These models may help identify patients suitable for direct intervention, bypassing MRCP.
• Further validation with prospective data is required to confirm clinical utility and refine accuracy.