Posted by
gankaku on
URL: http://imagej.273.s1.nabble.com/Help-with-quantifying-pictures-after-channel-splitting-tp5011445p5011544.html
Hi Stephanie,
A few general comments.
1.) When you say quantify the question which remains is quantify what
exactly. In your case it seems fairly clear that you want to measure the
area occupied by either your red staining (since you mentioned red is of
interest for you). Just to cite a lot of people mentioning this here once
in a while...Quantifying intensities in an image of histological stainings
cannot be done because of the missing linearity between darkness/strength
of the stain and the related amount of product/substrate.
2.) In your fist image you see that after application of the threshold in
the blue channel, the complete right hand side is considered by the
threshold. By looking at the image, this is rather unlikely to really be
red staining but rather a consequence of unequal lighting and will mislead
you in your measurements. Thus, I would first try to correct the lighting
at the microscope by setting it up with Koehler illumination (e.g. see:
http://zeiss-campus.magnet.fsu.edu/articles/basics/kohler.html) and if
available in your microscopic software acquire the image using "shading
correction". You can also afterwards correct for uneven illumination
(described here:
http://imagejdocu.tudor.lu/doku.php?id=howto%3Aworking%3Ahow_to_correct_background_illumination_in_brightfield_microscopy)
but this always bears the chance that you also introduce artifacts since
most post-priori methods are based on calculation except you collect a
flat-field image beforehand and use this for correction. You could try the
corrections in the BioVoxxel Toolbox which are two easy-to-use
implementation of the described flat-field corrections from the link above.
3.) When you save your images as .JPG you will always have to deal with
compression artifacts. Since (as it looks to me) your regions of interest
between the nuclei which show red staining are fairely small those
artifacts will increase the measurement error, markedly. Thus, I would try
to save the micrographs as tiffs or better in the original file-format of
the microscope. For most standard formats the BioFormat plugins for ImageJ
are able to read those.
4.) When you use the thresholding dialog and asjust the thresholds every
time manually you do the analysis only based on your visual perception.
Since human eyes are not very good in this, it is rather not recommendable
to use manual methods trying to achieve comparable results. Besides this,
you treat every image different (except if you use the same threshold every
time) and this would actually not allow a consecutive comparison. You could
use your original method, split the channels but thereafter apply the same
auto threshold to all your images. Since all your stainings have a certain
variability, this might account better for those and you work with the same
algorithm on the complete series of images.
5.) Besides the RGB channels I would also have a look into the HSB (>Image
>Type >HSB Stack) or Lab color space (>Image >Color >RGB to CIELAB). Those
often lead to an easier separation of distinct colors. You could also try
the "Color Threshold" for any of those color spaces by running the
threshold dialog on the original RGB image (this should start the color
thresholding dialog). Limitation here is that you again (see 5.) need to
define on representative images a threshold manually (!) but the advantage
would be that you can adjust this directly for RGB, HSB or Lab color space
and thereafter record it as a macro. The latter enables you to apply your
criteria to all your images which supports better comparability. However,
this might still suffer from not accounting for staining variabilities as
good as an auto threshold.
6.) If you can produce sample which show the blue and the red staining
separately (by e.g. applying the chemical staining components separately)
you could try to define vectors which separate your colors using the colour
deconvolution plugin (>Image >Color >Colour Deconvolution). This might also
reliably separate your staining efficiently and enable you to finally auto
threshold the individual extracted components.
There are other possibilities for sure but I think this info should be
sufficient for the beginning.
I hope this helps a little and is not too confusing ;-)
cheers,
Jan
2015-02-04 15:44 GMT+01:00 Stephanie Klett <
[hidden email]>:
> Dear Image J discussion group,
> I have a question about how I can quantify my digital microscopic images
> (.jpg) In my case I done a immunological staining where red is the colour
> of
> interest. Therefore, I split the channels and use the blue labelled picture
> and apply the function Threshold... (under Image-->Adjust-->). Then I get a
> histogram and an automatically adjustment of the colour sensibility (?). My
> question is what should I keep in mind when I want to compare several
> images. Do I have to adjust always the same sensibility (from the rage of 0
> to 255) or do I choose the automatically adjustment? My problem is, that
> sometimes the automatically adjustment represent the quantification better
> than the sensibility rate that I decided to use and other way around. Which
> way is more unbiased or do I have to do a completely other procedure?
>
> Here I display screen shots where I document my way and may problem more
> exemplified.
>
> Fig. 1 Opened a file and split the channels
>
> Fig. 2 Apply the function ³Threshold...² on the image labelled "blue³
>
> Fig. 3 Display the automatically adjustment of 228
>
> Fig. 4 Corrected the threshold to 180
>
> Fig. 5 A new image, display the automatically threshold of 222
>
> Fig. 6 Corrected the threshold to the undoing rate of 180
>
>
> I¹m very happy if you can help me and looking forward to your answer.
>
> Thank you for your attention
>
> Best regards
> Stephanie
>
> --
> ImageJ mailing list:
http://imagej.nih.gov/ij/list.html>
>
--
CEO: Dr. rer. nat. Jan Brocher
phone: +49 (0)6234 917 03 39
mobile: +49 (0)176 705 746 81
e-mail:
[hidden email]
info:
[hidden email]
inquiries:
[hidden email]
web: www.biovoxxel.de
--
ImageJ mailing list:
http://imagej.nih.gov/ij/list.html