macro threshold values

[Natalie M]

Hi all,
I am writing a macro to count cells in images I've taken of slides. Does
anyone know what code I could use to have the macro threshold each image at
the top 2% of each histogram, rather than at a set number value on each
histogram?
Thanks in advance,
Natalie Morel
University of Wisconsin

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

[CARL Philippe (LBP)]

Dear Natalie,
I guess you are looking for a code like this:
        nBins = 256;
        ratio = 0.5;
        run("Blobs (25K)");
        getHistogram(values, counts, nBins);
        Array.getStatistics(values, valuesMin, valuesMax, valuesMean, valuesStdDev);
        setThreshold(0, ratio * valuesMax );
        setOption("BlackBackground", true);
        run("Convert to Mask");
with in your case ratio = 0.98;
My best regards.
Philippe

Philippe CARL
Laboratoire de Biophotonique et Pharmacologie
UMR 7213 CNRS - Université de Strasbourg
Faculté de Pharmacie
74 route du Rhin
67401 ILLKIRCH
Tel : [phone]

-----Message d'origine-----
De : ImageJ Interest Group [mailto:[hidden email]] De la part de Natalie M
Envoyé : lundi 29 février 2016 18:43
À : [hidden email]
Objet : macro threshold values

Hi all,
I am writing a macro to count cells in images I've taken of slides. Does anyone know what code I could use to have the macro threshold each image at the top 2% of each histogram, rather than at a set number value on each histogram?
Thanks in advance,
Natalie Morel
University of Wisconsin

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

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

[Jeremy Adler]

Hi Natalie,

Do you want to select 2% of the total number pixels (those that are the most intense) or a population comprising of the most  intense pixels that make up 2% of the total intensity above background, or something else ?

This may not be a good approach for counting cells if the number of cells is variable.


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Natalie M
Sent: den 29 februari 2016 18:43
To: [hidden email]
Subject: macro threshold values

Hi all,
I am writing a macro to count cells in images I've taken of slides. Does anyone know what code I could use to have the macro threshold each image at the top 2% of each histogram, rather than at a set number value on each histogram?
Thanks in advance,
Natalie Morel
University of Wisconsin

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

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

[Swayne, Theresa C.]

Hi Natalie,


If you want to get the top 2% of the pixel population, there is a possibly related thread on the forum, where the poster is trying to get the top 2% of the pixel population using a cumulative sum.
http://forum.imagej.net/t/thresholding-as-a-percentage-of-intensity-histogram-in-a-macro/923/

However, I agree with Jeremy that differing amounts of positive staining per image will change the shape of the histogram, so the population approach may not be robust.

Philippe’s solution gives you a threshold = 98% of the maximum value in the image, which would seem to be robust to differing amounts of positive signal per image, as well as to inconsistent staining levels.

Hope this helps,
Theresa

On Mar 2, 2016, at 3:19 AM, Philippe CARL <[hidden email]<mailto:[hidden email]>> wrote:

Dear Natalie,
I guess you are looking for a code like this:
nBins = 256;
ratio = 0.5;
run("Blobs (25K)");
getHistogram(values, counts, nBins);
Array.getStatistics(values, valuesMin, valuesMax, valuesMean, valuesStdDev);
setThreshold(0, ratio * valuesMax );
setOption("BlackBackground", true);
run("Convert to Mask");
with in your case ratio = 0.98;
My best regards.
Philippe

Philippe CARL
Laboratoire de Biophotonique et Pharmacologie
UMR 7213 CNRS - Université de Strasbourg
Faculté de Pharmacie
74 route du Rhin
67401 ILLKIRCH
Tel : [phone]

-----Message d'origine-----
De : ImageJ Interest Group [mailto:[hidden email]] De la part de Natalie M
Envoyé : lundi 29 février 2016 18:43
À : [hidden email]<mailto:[hidden email]>
Objet : macro threshold values

Hi all,
I am writing a macro to count cells in images I've taken of slides. Does anyone know what code I could use to have the macro threshold each image at the top 2% of each histogram, rather than at a set number value on each histogram?
Thanks in advance,
Natalie Morel
University of Wisconsin

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

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

------------------------------------
Theresa Swayne, Ph.D.
Manager
Confocal and Specialized Microscopy Shared Resource<http://hiccc.columbia.edu/research/sharedresources/confocal>

Herbert Irving Comprehensive Cancer Center
Columbia University Medical Center
1130 St. Nicholas Ave., Room 222A
New York, NY 10032
Phone: [phone]
[hidden email]<mailto:[hidden email]>


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

[Ewing James]

From a statistics point of view, measures based on ranges (i.e., maximum values) are often not recommended because sample ranges depend on sample sizes.  You might be able to parameterize your distribution via its moments, and then take the predicted 98% upper confidence interval as a threshold.

- Jim

> On Mar 2, 2016, at 10:28 AM, Swayne, Theresa C. <[hidden email]> wrote:
>
> Hi Natalie,
>
>
> If you want to get the top 2% of the pixel population, there is a possibly related thread on the forum, where the poster is trying to get the top 2% of the pixel population using a cumulative sum.
> http://forum.imagej.net/t/thresholding-as-a-percentage-of-intensity-histogram-in-a-macro/923/
>
> However, I agree with Jeremy that differing amounts of positive staining per image will change the shape of the histogram, so the population approach may not be robust.
>
> Philippe’s solution gives you a threshold = 98% of the maximum value in the image, which would seem to be robust to differing amounts of positive signal per image, as well as to inconsistent staining levels.
>
> Hope this helps,
> Theresa
>
> On Mar 2, 2016, at 3:19 AM, Philippe CARL <[hidden email]<mailto:[hidden email]>> wrote:
>
> Dear Natalie,
> I guess you are looking for a code like this:
> nBins = 256;
> ratio = 0.5;
> run("Blobs (25K)");
> getHistogram(values, counts, nBins);
> Array.getStatistics(values, valuesMin, valuesMax, valuesMean, valuesStdDev);
> setThreshold(0, ratio * valuesMax );
> setOption("BlackBackground", true);
> run("Convert to Mask");
> with in your case ratio = 0.98;
> My best regards.
> Philippe
>
> Philippe CARL
> Laboratoire de Biophotonique et Pharmacologie
> UMR 7213 CNRS - Université de Strasbourg
> Faculté de Pharmacie
> 74 route du Rhin
> 67401 ILLKIRCH
> Tel : [phone]
>
> -----Message d'origine-----
> De : ImageJ Interest Group [mailto:[hidden email]] De la part de Natalie M
> Envoyé : lundi 29 février 2016 18:43
> À : [hidden email]<mailto:[hidden email]>
> Objet : macro threshold values
>
> Hi all,
> I am writing a macro to count cells in images I've taken of slides. Does anyone know what code I could use to have the macro threshold each image at the top 2% of each histogram, rather than at a set number value on each histogram?
> Thanks in advance,
> Natalie Morel
> University of Wisconsin
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> ------------------------------------
> Theresa Swayne, Ph.D.
> Manager
> Confocal and Specialized Microscopy Shared Resource<http://hiccc.columbia.edu/research/sharedresources/confocal>
>
> Herbert Irving Comprehensive Cancer Center
> Columbia University Medical Center
> 1130 St. Nicholas Ave., Room 222A
> New York, NY 10032
> Phone: [phone]
> [hidden email]<mailto:[hidden email]>
>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Natalie M]

Thank you all for your feedback! The reason I am thresholding the values to
count the cells is because the cells I want to count have been stained with
fluorescent dye, and I want to count only cells with the brightest
fluorescence (i.e., only the cells which are positive for the actin I am
measuring). Instead of counting them by hand, I thought setting a global
threshold would eliminate any bias on my part and allow me to choose which
cells are the most brightly stained in each image. Does this still seem
like an accurate way of achieving the cell count, or would something
different be recommended?
-Natalie

On Wed, Mar 2, 2016 at 9:53 AM, Ewing James <[hidden email]>
wrote:

> From a statistics point of view, measures based on ranges (i.e., maximum
> values) are often not recommended because sample ranges depend on sample
> sizes.  You might be able to parameterize your distribution via its
> moments, and then take the predicted 98% upper confidence interval as a
> threshold.
>
> - Jim
>
>
> > On Mar 2, 2016, at 10:28 AM, Swayne, Theresa C. <[hidden email]>
> wrote:
> >
> > Hi Natalie,
> >
> >
> > If you want to get the top 2% of the pixel population, there is a
> possibly related thread on the forum, where the poster is trying to get the
> top 2% of the pixel population using a cumulative sum.
> >
> http://forum.imagej.net/t/thresholding-as-a-percentage-of-intensity-histogram-in-a-macro/923/
> >
> > However, I agree with Jeremy that differing amounts of positive staining
> per image will change the shape of the histogram, so the population
> approach may not be robust.
> >
> > Philippe’s solution gives you a threshold = 98% of the maximum value in
> the image, which would seem to be robust to differing amounts of positive
> signal per image, as well as to inconsistent staining levels.
> >
> > Hope this helps,
> > Theresa
> >
> > On Mar 2, 2016, at 3:19 AM, Philippe CARL <[hidden email]
> <mailto:[hidden email]>> wrote:
> >
> > Dear Natalie,
> > I guess you are looking for a code like this:
> > nBins = 256;
> > ratio = 0.5;
> > run("Blobs (25K)");
> > getHistogram(values, counts, nBins);
> > Array.getStatistics(values, valuesMin, valuesMax, valuesMean,
> valuesStdDev);
> > setThreshold(0, ratio * valuesMax );
> > setOption("BlackBackground", true);
> > run("Convert to Mask");
> > with in your case ratio = 0.98;
> > My best regards.
> > Philippe
> >
> > Philippe CARL
> > Laboratoire de Biophotonique et Pharmacologie
> > UMR 7213 CNRS - Université de Strasbourg
> > Faculté de Pharmacie
> > 74 route du Rhin
> > 67401 ILLKIRCH
> > Tel : [phone]
> >
> > -----Message d'origine-----
> > De : ImageJ Interest Group [mailto:[hidden email]] De la part de
> Natalie M
> > Envoyé : lundi 29 février 2016 18:43
> > À : [hidden email]<mailto:[hidden email]>
> > Objet : macro threshold values
> >
> > Hi all,
> > I am writing a macro to count cells in images I've taken of slides. Does
> anyone know what code I could use to have the macro threshold each image at
> the top 2% of each histogram, rather than at a set number value on each
> histogram?
> > Thanks in advance,
> > Natalie Morel
> > University of Wisconsin
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> >
> > ------------------------------------
> > Theresa Swayne, Ph.D.
> > Manager
> > Confocal and Specialized Microscopy Shared Resource<
> http://hiccc.columbia.edu/research/sharedresources/confocal>
> >
> > Herbert Irving Comprehensive Cancer Center
> > Columbia University Medical Center
> > 1130 St. Nicholas Ave., Room 222A
> > New York, NY 10032
> > Phone: [phone]
> > [hidden email]<mailto:[hidden email]>
> >
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

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

[Swayne, Theresa C.]

Hi Natalie,

It’s absolutely right to try to set a threshold in an unbiased way.  It’s faster (in the long run) and more reproducible. The only difference folks are having is in how to find the most accurate method to select the threshold for this experiment.

You could use the same number for every image, or you could use a histogram-based method as you proposed, or you could use one of the automatic thresholding or trainable segmentation methods available in ImageJ.

The best approach will depend on your particular data —  e.g. difference between signal and background, variability of the positive staining level, and variability in positive cell numbers, from one image to the next.

In the end, you will need to test your method empirically on several representative images from different groups.

--Theresa

On Mar 2, 2016, at 2:15 PM, Natalie M <[hidden email]<mailto:[hidden email]>> wrote:

Thank you all for your feedback! The reason I am thresholding the values to
count the cells is because the cells I want to count have been stained with
fluorescent dye, and I want to count only cells with the brightest
fluorescence (i.e., only the cells which are positive for the actin I am
measuring). Instead of counting them by hand, I thought setting a global
threshold would eliminate any bias on my part and allow me to choose which
cells are the most brightly stained in each image. Does this still seem
like an accurate way of achieving the cell count, or would something
different be recommended?
-Natalie

On Wed, Mar 2, 2016 at 9:53 AM, Ewing James <[hidden email]<mailto:[hidden email]>>
wrote:

------------------------------------
Theresa Swayne, Ph.D.
Manager
Confocal and Specialized Microscopy Shared Resource<http://hiccc.columbia.edu/research/sharedresources/confocal>

Herbert Irving Comprehensive Cancer Center
Columbia University Medical Center
1130 St. Nicholas Ave., Room 222A
New York, NY 10032
Phone: [phone]
[hidden email]<mailto:[hidden email]>


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