Figure 3.

The framework of our method. We train the network end-to-end to ensure the best performance. The framework contains 3 sections: Feature Extraction Network (FEN), Region Proposal Network (RPN), and Class Prediction Network (CPN). The will help to extract the deep-level features of the image with the help of the relation module and Spatial Pyramid Pooling (SPP) layer, which is the input to RPN and CPN. The RPN will locate the position of essential structures based on the anchors generated by Feature Pyramid Network (FPN), and the CPN will help to judge and classify the structures. The final output will be a quality assessment of each essential structure and its location.