Machine learning-based diagnostic modeling for differentiating lymphoid hyperplasia from acute appendicitis using laboratory biomarkers.

BACKGROUND

This study aimed to assess the diagnostic ability of routine laboratory biomarkers and develop machine learning (ML) models to improve differentiation between LH and AA.

METHODS

A total of 873 patients (209 LH; 664 AA) were retrospectively analyzed. Laboratory parameters, including CRP, WBC, neutrophils, lymphocytes, monocytes, NLR (neutrophil to lymphocyte ratio), LMR (lymphocyte to monocyte ratio), PLR (platelet to lymphocyte ratio), and PIV (pan immune inflammation value), were used to build logistic regression, Naive Bayes, neural network, and gradient boosting models. Diagnostic performance was evaluated using AUC, accuracy, precision, recall, and F1 score.

RESULTS

All biomarkers differed significantly between LH and AA (p < 0.001). AA patients exhibited higher CRP, WBC, neutrophil count, NLR, PLR, and PIV, whereas LH demonstrated higher LMR values. Logistic regression yielded the best performance (AUC 0.918; accuracy 0.869), followed closely by Naive Bayes (AUC 0.917) and neural network (AUC 0.914). SHAP-based model interpretability analysis further identified LMR and NLR as the most influential features driving model predictions.

CONCLUSION

Routine hematologic biomarkers combined with ML-based modeling provide a robust, non-invasive tool for differentiating LH from AA.