Fig. 7: Generation of synthetic images and labels using GAN-based methods.
From: Improving 3D deep learning segmentation with biophysically motivated cell synthesis
The simulated cell borders (A) are converted into binary membrane signals (B), which are then utilized by the generator to produce synthetic nuclei signals (C) along with corresponding binary labels (D). Subsequently, a post-processing step is employed to extract instance labels (E) from the generated binary labels. The binary membrane signals (B) partially show large white regions. These are a result of the lower resolution in the z-direction, leading to cell borders/membranes that completely lie in one z-plane. Scale bar: 50 μm.
