Figure 2.

Illustration of ROI annotation in CEUS videos and the design of Radiomics and 3D-Resnet-50 models. (A) CEUS examinations were performed for each breast tumor. (B) An example of the yellow bounding box ROI drawn in one CEUS frame. (C) Schematic of radiomics combined with the XGBoost model. (D) A three-dimensional ROI (2D in space and 1D in time) of CEUS videos was fed into the 3D-Resnet-50 model to obtain the discriminative features by automatic feature learning. (E) The features obtained from the radiomics and 3D-Resnet-50 models are used to calculate the prediction probability. ROI, region of interest; CEUS, contrast-enhanced ultrasound; ResNet, residual network.