Related Experiment Videos
Development and external validation of a machine learning model for predicting gangrenous cholecystitis
Xianyue Ren1, Liang Wang1, Yue Zhu1
1Department of General Surgery, The First Affiliated Hospital of Jinzhou Medical University, Jinzhou, Liaoning, China.
Background:
Gangrenous cholecystitis (GC) is a severe surgical emergency with high postoperative morbidity and mortality, making early and accurate diagnosis essential for timely intervention. Conventional statistical models are limited in capturing nonlinear interactions among clinical variables and often lack interpretability, hindering individualized risk stratification. Machine learning enables accurate and interpretable risk prediction.
Methods:
The retrospective clinical study included a total of 492 patients with 39 routine serological tests and imaging tests collected. Seven models including LR, DT, RF, XGBoost, LightGBM, SVM and ANN were constructed. Model performance was evaluated using area under the curve (AUC), accuracy, sensitivity, specificity, and other metrics in train data, internal validation data and an independent external test data. An interpretable web-based risk calculator was dveloped based on the optimal model, with decile stratification employed to verify the reliability of predicted probabilities. SHapley Additive exPlanations (SHAP) were applied for feature interpretability.
Results:
The ANN model demonstrated the best performance, with an AUC of 0.932 (95% CI 0.900-0.961) in train data, 0.921 (95% CI 0.862-0.966) in internal validation data, and 0.867 (95% CI 0.790-0.929) in external validation. Five core predictors: NLR, WBC, FIB, NEU. and GBWT were identified as the most influential predictors. SHAP analysis provided interpretable insights into their contributions to infection risk. ANN model has been translated into a convenient tool to facilitate its utility in clinical settings.
Conclusion:
This study developed and externally validated an ANN-based prediction model for the early diagnosis of GC, achieving robust predictive performance and generalizability. The model holds promise for providing an early, personalized assessment of whether patients with cholecystitis will develop gangrene. Future prospective studies are warranted to further validate its efficacy.