Fig. 1
From: A deep error correction network for compressed sensing MRI
The proposed Deep Error Correction Network (DECN) architecture consists of three modules: a guide module, an error correction module, and a data fidelity module. The input of the error correction module is the concatenation of the zero-filled compressed MR samples and guidance image while the corresponding training label is the reconstruction error â–³xp. After the error correction module is trained, the guidance image and feed-forward approximation of the reconstruction error for a test image are used to produce the final reconstructed MRI