measuring colors-heterogenous areas

[Attin, Mina]

Hi,
I recently started to use ImageJ to quantify the staining (blue color) of
different heart tissue segments. I use a method that does not allow me to
quantify the % of staining in each segment accurately because the staining
(blue color) is very heterogeneous. The color of heart tissue is dark pink
and the staining is the blue color that is distributed heterogeneously
throughout each segment. I will explain my method and I will appreciate if
you have any suggestion to improve the method or suggest different way to
quantify the blue color. For each image, I start with image/color/RBG
split, then I use plugins to use filter/multithreshold. I try to determine
a reasonable threshold (the red color to cover the staining only-this is
hard because the staining is very heterogenous) and I measure the area.
Finally, I measure the total area of each segment and I will obtain the %
of staining by dividing the staining area to the whole area.

Best regards,

M

[Toby Cornish]

Dear Mina:

As I understand it, you are trying to segment an RGB color image of tissue by stain color.

It sounds like you are using thresholding in the R, G, and B channels to segment the image (RGB colorspace segmentation). Unfortunately, using the RGB colorspace to segment RGB images doesn't work well.  Two commonly used approaches are HSB colorspace segmentation or color deconvolution.  These are distinct approaches to the problem and the subsequent analyses of the resultant images are also different.  I have covered this before in this forum, so if you search the archives, you may turn up a better explanation.  If you consult wikipedia, you will find very good information about the HSB colorspace.  Ruifrok published a the paper about color deconvolution for histochemistry.

Gabriel Landini has excellent ImageJ plugins for both color deconvolution and HSB colorspace segmentation:
 
  http://www.dentistry.bham.ac.uk/landinig/software/software.html

We also have an ImageJ-based program for analyzing images using HSB colorspace segmentation.  It is possible to use this software to perform analyses of the sort you describe.  It is free for non-commercial use:

  http://bui2.win.ad.jhu.edu/frida/

If you have any additional questions, feel free to contact me.

Toby

> Date:    Tue, 9 Oct 2007 12:44:22 -0500
> From:    "Attin, Mina" <[hidden email]>
> Subject: measuring colors-heterogenous areas
>
> Hi,
> I recently started to use ImageJ to quantify the staining (blue color) of
> different heart tissue segments. I use a method that does not allow me to
> quantify the % of staining in each segment accurately because the staining
> (blue color) is very heterogeneous. The color of heart tissue is dark pink
> and the staining is the blue color that is distributed heterogeneously
> throughout each segment. I will explain my method and I will appreciate if
> you have any suggestion to improve the method or suggest different way to
> quantify the blue color. For each image, I start with image/color/RBG
> split, then I use plugins to use filter/multithreshold. I try to determine
> a reasonable threshold (the red color to cover the staining only-this is
> hard because the staining is very heterogenous) and I measure the area.
> Finally, I measure the total area of each segment and I will obtain the %
> of staining by dividing the staining area to the whole area.
>
> Best regards,
>
> M

Toby C. Cornish, M.D., Ph.D.
Pathology Resident
Johns Hopkins Medical Institutions
[hidden email]

[Jacqueline Ross]

Dear Mina,

If you are looking for any further instructions on colour deconvolution,
you are welcome to have a look at some notes on colour analysis that I
put together for research staff/students here.

You can access the notes at:
http://www.health.auckland.ac.nz/biru/web_resources/image_analysis_notes
.html

The PDF for the colour analysis is quite large so it will take a while
to download.

By the way, if anyone finds any errors in them, please let me know!

Kind regards,

Jacqui

Jacqueline Ross
Biomedical Imaging Research Unit
School of Medical Sciences
Faculty of Medical & Health Sciences
The University of Auckland
Private Bag 92019
Auckland, NEW ZEALAND

Tel: [phone] Ext 87438
Fax: [phone]

http://www.health.auckland.ac.nz/biru/ 
-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
Toby Cornish
Sent: 11 October 2007 07:25
To: [hidden email]
Subject: Re: measuring colors-heterogenous areas

Dear Mina:

As I understand it, you are trying to segment an RGB color image of
tissue by stain color.

It sounds like you are using thresholding in the R, G, and B channels to
segment the image (RGB colorspace segmentation). Unfortunately, using
the RGB colorspace to segment RGB images doesn't work well.  Two
commonly used approaches are HSB colorspace segmentation or color
deconvolution.  These are distinct approaches to the problem and the
subsequent analyses of the resultant images are also different.  I have
covered this before in this forum, so if you search the archives, you
may turn up a better explanation.  If you consult wikipedia, you will
find very good information about the HSB colorspace.  Ruifrok published
a the paper about color deconvolution for histochemistry.

Gabriel Landini has excellent ImageJ plugins for both color
deconvolution and HSB colorspace segmentation:
 
  http://www.dentistry.bham.ac.uk/landinig/software/software.html

We also have an ImageJ-based program for analyzing images using HSB
colorspace segmentation.  It is possible to use this software to perform
analyses of the sort you describe.  It is free for non-commercial use:

  http://bui2.win.ad.jhu.edu/frida/

If you have any additional questions, feel free to contact me.

Toby




> Date:    Tue, 9 Oct 2007 12:44:22 -0500
> From:    "Attin, Mina" <[hidden email]>
> Subject: measuring colors-heterogenous areas
>
> Hi,
> I recently started to use ImageJ to quantify the staining (blue color)
of
> different heart tissue segments. I use a method that does not allow me
to
> quantify the % of staining in each segment accurately because the
staining
> (blue color) is very heterogeneous. The color of heart tissue is dark
pink
> and the staining is the blue color that is distributed heterogeneously
> throughout each segment. I will explain my method and I will
appreciate if
> you have any suggestion to improve the method or suggest different way
to
> quantify the blue color. For each image, I start with image/color/RBG
> split, then I use plugins to use filter/multithreshold. I try to
determine
> a reasonable threshold (the red color to cover the staining only-this
is
> hard because the staining is very heterogenous) and I measure the
area.
> Finally, I measure the total area of each segment and I will obtain
the %
> of staining by dividing the staining area to the whole area.
>
> Best regards,
>
> M



Toby C. Cornish, M.D., Ph.D.
Pathology Resident
Johns Hopkins Medical Institutions
[hidden email]