Subject: | |
From: | |
Reply To: | |
Date: | Wed, 28 Feb 2001 20:33:02 +0200 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
Kees Jalink wrote:
>
>
> But before trying to track down all those pieces of software to run our
> data through, what is the basis of quantification? More specifically,
> what statistical (or other) math is used to come to a number that
> describes just how well 2 images colocalize? I have tried a bit using
You raise a key point -- ie, when is the number significant.
One idea that I am trying is to ask what overlap would be expected if
the labeled areas were randomly placed in a given area. To do this I
mask the labeled areas -- that is, I do not weight them as is done in
the Manders coefficients. This may not be the best model, but it is
solvable, and it also behaves reasonably. That is , if you have a high
percent of your area covered with a label, then you need an extremely
high correlation coefficient
to claim colocalization. But if you only have a small percent coverage,
then you can say there is colocalization even with a much lower
correlation coefficient.
--aryeh
--
Aryeh Weiss | email: [log in to unmask]
Department of Electronics | URL:
http://optics.jct.ac.il/~aryeh
Jerusalem College of Technology | phone: 972-2-6751146
POB 16031 | FAX: 972-2-6751275
Jerusalem, Israel | ham radio: 4X1PB/KA1PB
|
|
|