Based on a radiomics-clinical nomogram to predict the therapeutic effect of pulmonary cryptococcosis pneumonia.

We aimed to develop a radiomics-clinical nomogram to predict the therapeutic effect of PC pneumonia. A total of 255 PC pneumonia patients (165 cases with good therapeutic effect and 90 cases with poor therapeutic effect) were retrospectively enrolled from two centers. 190 PC pneumonia patients from Center 1 were randomly divided into a training group (133 cases) and an internal validation group (57 cases) at a ratio of 7:3. The data from Center 2 were used as the external validation group (65 cases). The radiomics features selected from unenhanced CT were input into multiple machine learning models. The optimal model was obtained through receiver operating characteristic (ROC) analysis to construct the radiomics and clinical models. Combined them to build a nomogram for clinical application. These models were evaluated by ROC analysis, calibration curve, and decision curve analysis (DCA). The support vector machine was the best classifier with the highest area under the curve (AUC) of 0.848 in the internal validation group. Although the performance of the nomogram was similar to that of the radiomics model (AUC: 0.886 vs. 0.864; Delong test: P = 0.632), they were significantly higher than that of the clinical model (AUC: 0.886 vs. 0.730, 0.864 vs. 0.730; Delong test, P < 0.05) in the external validation group. The calibration curves and DCAs of the nomogram and radiomics models confirmed their superior predictive agreements and clinical utilities. Radiomics model and nomogram could noninvasively predict the therapeutic effect of PC pneumonia, which was beneficial for precise intervention.