Machine learning model for differentiating xanthogranulomatous cholecystitis and gallbladder cancer in multicenter...

Area of Science:

• Hepatobiliary surgery
• Medical artificial intelligence
• Diagnostic imaging analysis

Background:

• Differentiating xanthogranulomatous cholecystitis (XGC) from gallbladder cancer (GBC) preoperatively is clinically challenging due to similar presentations.
• Accurate preoperative diagnosis is crucial for appropriate treatment planning and patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model (LIDGAX) for the preoperative differentiation of XGC and GBC.
• To assess the diagnostic performance and clinical utility of LIDGAX compared to existing methods and expert radiologists.

Main Methods:

• A multicenter retrospective study involving 1246 patients (554 XGC, 692 GBC).
• Development of the LIDGAX ML model using preoperative clinical, imaging, and laboratory data.
• Multivariate logistic regression and least absolute shrinkage and selection operator analyses identified key predictive variables.
• Internal and external validation, comparison with other ML models and radiologists, and calibration/decision curve analyses were performed.

Main Results:

• LIDGAX achieved high performance with AUC values of 0.94 (internal) and 0.88 (external validation).
• The model outperformed five other ML models and demonstrated superior clinical utility.
• LIDGAX improved sensitivity, specificity, and balanced accuracy compared to radiologists, while significantly reducing diagnostic time.

Conclusions:

• The developed LIDGAX model offers a non-invasive and accurate tool for preoperative differentiation of XGC and GBC.
• LIDGAX shows strong potential for clinical translation, enhancing diagnostic capabilities in hepatobiliary surgery.
• The open-source deployment of LIDGAX ensures accessibility and continued high performance for clinical application.