Fig. 3.

Image-to-image translation example with our proposed approach. The figure shows a paired example of a real non-fat-saturated and fat-saturated MR knee image with the reconstructed MR image (prediction). The images are annotated with the true acquisition parameters (TE, TR) for the input and target, and with the acquisition parameters determined by the AC for the predicted image. Both models generate sharp MR images with the correct contrast (see TR/TE values). However, model 6 produces inaccurate results and contrast, e.g., in the bone structures