Supplementary Figure 6: Comparison of ground-truth (“target”) and predicted pixel values in images from Supplementary Fig. 2.
From: Label-free prediction of three-dimensional fluorescence images from transmitted-light microscopy
For each structure indicated, 1% of pixels were randomly selected from the target image are plotted against the corresponding pixels in the predicted image. The number of pixels (n) compared was 36,126 for the microtubules prediction, 96,501 for the cell membrane prediction, 96,501 for the DIC nuclear envelope prediction, and 48,168 for all other predictions. Pixel intensities are normalized in the target in predicted images (see Methods).
