An interpretable machine learning framework for pediatric appendicitis and peritonitis diagnosis

Appendicitis is a predominant condition in which the appendix ruptures due to bacterial infection. This rupture later leads to critical and life-threatening complications such as peritonitis, sepsis, abdominal pain, and septic thrombophlebitis. In this paper, we present a dual-target machine learning framework for presumptive diagnosis and peritonitis prediction using structured data from clinics and laboratories. The initial clinical judgment made before the confirmatory tests is known as a presumptive diagnosis, while peritonitis is a serious inflammation of the peritoneum. The Boruta algorithm was utilized to analyze the important appendicitis markers. Seven different Explainable artificial intelligence (XAI) procedures, such as Shapley Additive Values (SHAP), Q-Lattice, Local Interpretable Model-Agnostic Explanations (LIME), Partial Dependence Plots (PDP), Individual Conditional Expectation (ICE), Counterfactual Explanations and Anchor, were utilized to improve interpretability and provide transparency.The findings demonstrate that combining advanced data resampling methods such as Borderline-SMOTE with optimized machine learning classifiers such as soft-voting configuration yields best performance with accuracy of 93.24% and an AUC of 0.99 for presumptive appendicitis diagnosis indicating improved predictive accuracy and strong greater robustness.Explainable analyses identified key clinical indicators,including appendix diameter,white blood cell count,Alvarado score,red cell distribution width and length of stay as the most influential predictors across multiple XAI frameworks.As a supplementary deployment-oriented assessment, ChaCha20-Poly1305 authenticated encryption was evaluated on the pediatric clinical dataset to examine confidentiality, integrity, and computational efficiency requirements for secure clinical data storage and transmission supporting robust protection of clinical workflow.Overall this study illustrates how machine learning and explainability can be utilized to improve peritonitis and appendicitis early discovery in medical settings.