Skip to main content
. Author manuscript; available in PMC: 2023 Sep 1.
Published in final edited form as: Neuroimage. 2021 Sep 8;243:118569. doi: 10.1016/j.neuroimage.2021.118569

Fig. 5.

Fig. 5.

Training the pseudo 3D implementation includes two parts: (a) the harmonization network is first trained using a pool of multi-site multi-orientation MR slices. Then (b) weights of the harmonization network are frozen to train the 3D fusion network. The fusion network learns to combine the three stacked volume images into one single fused image.