Fig. 3

Performance evaluation of the ML-based biomarker panel for all-stage PDA diagnosis. (a) ROC curve comparing the diagnostic performance of multiple ML models in cohort A. (b) Confusion matrices for different ML algorithms. (c) Feature importance analysis for all-stage PDA using Random Forest. (d) Feature importance analysis for all-stage PDA using XGBoost. (e) Feature importance analysis for all-stage PDA using LightGBM. (f) Feature importance analysis for all-stage PDA using CatBoost.