Re: Only count cells that contain nuclei (or holes in the binarized image)

Posted by Gabriel Landini on Oct 16, 2015; 8:53am
URL: http://imagej.273.s1.nabble.com/Only-count-cells-that-contain-nuclei-or-holes-in-the-binarized-image-tp5014654p5014658.html

There is a further way of computing those relations using region connection
calculus:
http://www.mecourse.com/landinig/software/spatial/rcc8d.html

See the references listed in that page for specific examples. You perhaps are
interested in the relations of RCC8D logic that describe TPP(nucleus, cell)
(for nuclei that touch the cell boundary), and NTPP(nucleus, cell) (for nuclei
that are inside the cells but not touching the boundary), or the relation
PP(nucleus, cell) in the RCC5D logic (both are supported by the plugins).

However, I think the image you posted will probably not return accurate
results (with any method) due to many cells appearing to be merged into
clusters and also there are many small areas which are not cells and many
small nuceli which are not strictly so.

Before doing further processing you will benefit from cleaning the images of
those small regions and holes and separate the touching cells. You might want
to try opening and closing, or median filtering.
 
You could use a region labelling plugin to see what are the connected regions
and check if they are one or multiple cells, See the attached image, clusters
with the same colour will count as "one cell" but are in fact many.

The watershed separation might be useful to separate those clustered cells,
but you will have to fill the holes first.
Hope it helps.

Gabriel


--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html