Figure 3.

Performance of the deep learning model for the differentiation of MIBC and NMIBC. (A) Receiver operator characteristic curves of the model in four different cohorts. (B) Comparison of the performance between the model and two radiologists. (C) Calibration curves of the model in internal and external validation cohorts. The calibration curve showed that the predicted probabilities generally agreed with the observed probabilities. The predictive performance of the model in the external validation cohort exhibited a closer fit to the perfect calibration. (D, E) showed decision curve analyses (DCA) in the internal and external validation cohorts respectively. DCA compared the net benefit of the deep learning model versus treat all or treat none are shown. The net benefit was plotted versus the threshold probability. The net benefits of the deep learning model (blue line) were superior to the benefits of treating all or treating none.