Deep Learning Combined with Radiologist’s Intervention Achieves Accurate Segmentation of Hepatocellular Carcinoma in Dual-Phase Magnetic Resonance Images

Figure 3

Architecture of the proposed deep fusion network (DFN). DFN includes 2 components: subdeep convolutional neural networks (DCNN) and fusion blocks. The input data initially go through different sub-DCNNs, and a fusion block gathers their segmentation score maps and integrates them to reconstruct the final delineation outcome.