Figure 3

Study workflow. Three models were developed in the training dataset. First, the CT model and the radiomics model were developed. The better one was chosen, and its output, named the “Rad-score”, was taken to join the clinical information for the development of the combined model. Then, the three models were evaluated in the test dataset. ROC and DCA analyses were performed, and the best model was chosen for further evaluation with precision-recall curve and calibration curve analyses. Finally, a nomogram was developed as the graphical representation of our best model.