You can try to use first Process >Find Maxima, optimize the point count
using the noise tolerance an choosing Maxima with tolerance as an output.
This might give you binary particles which you can use with any neareat
neighbor plugin you tried so far. If the specified areas ate completely odd
and your cells are rather homogenious in size you could try "Ultimate
Points " as output and threshold those points with a manual threshold set
to 1 and 255. Then continue with the nearest neighbor plugin.
But be aware that your object has a curvature which will influence the
accuracy of your analysis