Skip to main content
. 2020 Nov 5;16(1):81–89. doi: 10.1007/s11548-020-02275-z

Fig. 3.

Fig. 3

Illustration of 3D Glow-based data augmentation. Two training VOIs, xa and xb, are mapped to points za and zb of the latent space, which are used for interpolating a new point z~. By mapping z~ back to the VOI space, we generate a new synthetic VOI, x~, which is different from, yet similar to xa and xb