2.4.3. T‐distributed stochastic neighbor embedding of style‐codes

To illustrate whether the 1 × 64 style code was successfully injected into the harmonized images, a t‐distributed stochastic neighbor embedding (t‐SNE) plot (Van der Maaten & Hinton, 2008) was used to visualize the style representations of images randomly selected from the ADNI, UKBB and PPMI datasets, respectively. Briefly, t‐SNE is a nonlinear dimensionality reduction method for visualizing high dimensional data, where more similar data points are closer together, and dissimilar points further apart. The style code was extracted from the style encoder trained in the model before and after the images were harmonized.