Performance of Deep Learning-Based Algorithm for Detection of Pediatric Intussusception on Abdominal Ultrasound Images

Figure 3

Network structure. The main structure of the network consists of three stages: backbone for feature extraction network, region proposal network (RPN), and region of interest (ROI) pooling. Input the image into the VGG16 to get the feature map, use RPN to generate the anchors, after Nonmaximum Suppression (NMS) to obtain ROIs, project the ROIs onto the feature map to get the feature matrix, scale each feature matrix to through the ROI pooling, and then flatten the feature map to get the prediction result through a series of fully connected layers.