2.5. Machine learning-based segmentation and classification
ilastik was used to distinguish the background from the yeast to be
counted. For example, the pixel classification function of the software
was used to remove the background of the compressed pictures in batches.
The background and yeast samples can be marked many times by different
labels under the Training→Live update command in the process of
background removal. Regardless of whether the background was simple or
complex, it was separated from the yeast to be counted as much as
possible (Figure 1C and Fig.SC).
2.6. Establishment and application of ImageJ macro“Yeast Counter”
The ImageJ macro “Yeast Counter” was written in a Java-like
programming language, and full operational details are provided in the
“Yeast Counter” in the supplementary material. We smoothed and
adjusted the threshold of the images processed in batches using ilastik
and ImageJ (Image→Adjust→Threshold→Li→Process→Smooth→apply), and
subsequently black-and-white images were presented. Then, the
Process→Binary→Fill hole and Watershed functions in ImageJ were used to
fill the gap (the part marked by the red circle in Figure 1D, Fig.SD)
and separate joined cells (the position indicated by the red clipper in
Figure 1E, Fig.SE). In addition, Analyse→Analyse Particles → Size, Area,
Circ and other indexes can be used to assess the objectives in images.
Here, only the Area command was applied to distinguish yeast cells from
background fragments (Figure 1F and Fig.SF). Then,
the
threshold in the Analyse Particles command was set, and the number of
yeast cells was counted automatically (Figure 1G and Fig.SG, ”count.ijm”
in the supplementary material). The final results were saved as a file
in csv format.