Figure 3.

Framework of the deep learning network used in this study. (A) Framework used in single-phase models was implemented in the pre-contrast, arterial, portal, and hepatobiliary phase models. (B) Framework of attention-guided feature fusion network used for the pre-contrast + arterial + portal + hepatobiliary phase model (4P). The arterial + portal phase model (2P) and the arterial + portal + hepatobiliary phase model (3P) used similar frameworks. The self-attention block and collaborative-attention block were added between the convolutional and pooling layers.