Counting Cells in sliced microscopic view

At a research lab, we currently have students manually count the number of cells after a tissue sample is sliced and observed under microscope.
The problem here is that the cells are well inter-connected, even after turning the image in grayscale. Tried to apply a manually specified binary threshold (because wasn’t able to get OTSU to find an optical threshold value), and attempted to apply watershed to the image, followed by finding contours. But so far no dice. Does anyone have a suggestion as to how to best automate this cell counting process?

Problem 1: Cells are pretty next to each other, closely
Problem 2: Difficulty in finding optimal threshold value for binary separation
Problem 3: Cells are not circular in shape, thus, Huges circles aren’t working well here
Problem 4: Cells are of no particular size. Some are big…are are small… in irregular shapes.

Appreciate any feedbacks and comments!


It would be much easier to count if the cells were separated. The most straightforward way to do this is would be to add a marker that marks a singular, more easily separable feature of the cell, like the nucleus. Can you apply a simple H&E stain which marks the nucleus blue?

Otherwise, it seems hard to count by eye, so I wouldn’t expect CellProfiler to do any better than manual counting.