Accurately counting objects touching borders

Hi,
What would be the cleanest procedure for counting objects that are touching borders in the image?

Simply including them for each field image could lead to the same object being counted in overlapping field images. I am considering stitching together all the fields within a well prior to running my CP pipeline. However, I can have as many as 100 field images per well, each 500 kb in size, so 50 MB in total per well.

Are there any considerations I should take into account before trying this? Will CP run much slower with one giant well image instead of all the small images?

One reason behind why I haven’t tried this yet, is that our microscope arranges fields in a spiral starting from the center of the well and I can’t seem to find an option to tile images according to this pattern in CP. Am I just missing it or does CP only tile by rows and columns at the moment? If so, I will try to write my own script to stitch them together.

Will CP run much slower with one giant well image instead of all the small images?

Yes, but 50MB still isn’t that big as images go, so I think that’d be handled by CP still at reasonable speed. There’s a new software I haven’t personally tried called Orbit that’s designed for big tissue images and integrates with CP, so that’s another option to try.

Am I just missing it or does CP only tile by rows and columns at the moment?

CP only tiles by rows and columns; more sophisticated layouts aren’t supported. I’d personally try ImageJ’s stitching plugin, while I’ve not tried this option myself it says that it supports a user-defined image map.

What would be the cleanest procedure for counting objects that are touching borders in the image?

If you can stitch your images cleanly and if CP can handle them, I think analyzing the stitched images would be best. If one or both of those isn’t true, you can try allowing objects to touch the borders but then implementing a size filter that’s say 60% the size of an ‘average’ cell; it means you may under-count very small cells touching the edges or over-count very big ones, but on the whole I’d think it’d give you a reasonable accuracy.

1 Like