Fig. 1

The overall pipeline of this study. (a) Flowchart of patient recruitment, the cohort of patients in our hospital was randomly divided into training set, validation set, and internal-test set, 88 patients from another hospital were assigned to a separate external-test set; (b) construction of five different DL models for identifying BPT and MPT based on US images; (c) comparison of diagnostic performance among different models using the AUC to select the best model in internal- and external- test set; (d) evaluation of whether radiologists with varying levels of experience show improved diagnostic performance with the assistance of the model