Fig. 10.

Results of domain adaptation. The proposed model is trained on Sites $A$ and $D$, and the task is to harmonize images from Site $C$ to Site $D$ during testing. From left to right: (a) the original image from Site $C$; (b) harmonization without domain adaptation; (c) harmonization with domain adaptation; (d) harmonization after a retraining that includes Sites $A$, $C$, and $D$; (e) target image (the same subject imaged at Site $D$). Red arrow indicates a bad harmonization in the cerebellum region without domain adaptation.