Re: Cell counting with imageJ

Posted by Gabriel Landini on
URL: http://imagej.273.s1.nabble.com/Cell-counting-with-imageJ-tp3690375p3690378.html

On Sunday 22 November 2009, Lukas Hoffmann wrote:

> You're right, I never thought of that.  For the sample images, users could
>  mark the cells' locations with the ImageJ manual cell counter plugin.
>  Then it runs the tests.  If there are multiple particles close to the
>  marker (like when cell is fragmented into 10 pieces), those are scored as
>  a duplicate count.  If a particle is not close to a marker, that is scored
>  as a false count.  Closeness is defined as 'inside the user-set cell
>  radius'.  So the tests should measure # of particles counted, number of
>  duplicate counts, number of false counts.  It would choose the image
>  processing algorithm that optimizes all three.  The user would not have
>  extra work since he has to do some manual counting anyway.

I think that more important is to look into how well do the segmented pixel
groups match the expected segmentation (i.e. a ground-truth or gold standard
image).
If I understand the above correctly one risks a single noise pixel very near a
non-detected cell to be accepted as correct.

Have a look here for some interesting ideas (the last part of the paper):
Sezgin, M & Sankur, B (2004), "Survey over Image Thresholding Techniques and
Quantitative Performance Evaluation", Journal of Electronic Imaging 13(1):
146-165.

Cheers

G.