Fig. 2. The performance of the ensembled deep learning model, human experts, and the combinations of model and humans for the diagnosis of biliary atresia on the external validation dataset.

a The ROC curve of the model at the image level. b The ROC curve of the model at the patient level based on majority vote. c The ROC curve of the model at the patient level based on single image with best image quality for each patient. d The performance of the combined deep learning model and human expert (circles) at the image level. The blue star represents the performance of the model with the default threshold (0.5) to binarize outputs of the model. Source data are provided as a Source data file. ROC receiver operating characteristic.