Table 1 Synthesis performance of the PACGAN for the final model trained on the entire training set, computed considering the images of the test set as the “real” distribution. The quality of the synthetic images was evaluated using the Fréchet inception distance (FID), kernel inception distance (KID), and structural similarity index measure (SSIM). The range of each metric is indicated in the header. The symbol \(\:\uparrow\:\) indicates that a higher value corresponds to greater similarity, while \(\:\downarrow\:\) means the opposite. The quality metrics computed for the comparison between real and synthetic images (Real/Fake) serve as an index of the realness of the synthetic images. The same metrics computed between images from the same distribution (Real/Real and Fake/Fake) enable the assessment of the diversity of the generated images. Image data IDs for real images used in the computation are provided in supplementary methods, and synthetic images are accessible as a NIfTI file in the supplementary material.

From: High-resolution conditional MR image synthesis through the PACGAN framework