Fig. 2: Model performance evaluation.
From: Deep learning radiomics based prediction of axillary lymph node metastasis in breast cancer
a–c Receiver operating characteristic area under the curves for the proposed clinical model and DLRN in EVC1, EVC2, and EVC3, respectively. d–f Decision curves of the clinical model and DLRN in EVC1, EVC2, and EVC3, respectively. g–i Calibration curves of DLRN in EVC1, EVC2, and EVC3, respectively. EVC, external validation cohort; CLI, clinical model; DLRN, deep learning radiomics nomogram. Note: In Fig. 2d–f, the purple and blue lines represent the DLRN and clinical model, respectively. The orange line represents the assumption that all cases underwent ALN dissection or SLN biopsy. The green line represents the assumption that no cases underwent ALN dissection or SLN biopsy. The decision curve reveals that the DLRN exhibits better performance than the clinical model over a wide range of threshold probabilities.
