Fig. 4

Calibration and DCA for machine learning models. (A) Calibration curves for different models, showing the relationship between predicted probabilities (x-axis) and actual positive rates (y-axis). The dashed line represents perfect calibration. The Brier score (in parentheses) indicates calibration performance, with lower values indicating better calibration. Logistic Regression has the lowest Brier score (0.1496), indicating the best calibration. (B) DCA for different models. The x-axis represents the decision threshold probability, and the y-axis shows net benefit. “Treat All” (red line) and “Treat None” (black line) are the reference strategies. Logistic Regression, SVM, and DNN show the highest net benefits, indicating superior clinical utility across various thresholds. *DCA: Decision curve analysis.