Journal: NPJ digital medicine
This multicenter retrospective study developed a machine learning model, LIDGAX, to preoperatively differentiate xanthogranulomatous cholecystitis (XGC) from gallbladder cancer (GBC). This differentiation is challenging due to overlapping clinical and imaging features.
Using data from 1,246 patients, twelve key predictive variables were identified. LIDGAX demonstrated strong diagnostic performance, including:
- AUC of 0.94 in internal validation
- AUC of 0.88 in external testing
- Outperformance of five other machine learning models
The model also showed improved sensitivity, specificity, and balanced accuracy compared to six radiologists, while reducing diagnostic time per patient by about 30 seconds.
LIDGAX was successfully deployed on an open-source online platform, maintaining high accuracy (AUC 0.95, accuracy 0.92).
This non-invasive tool shows promise for clinical use in preoperatively distinguishing XGC from GBC.