Segmentation and analysis of silver grains in TEM image

[Jacqueline Ross]

Dear All,

 

I recently had an enquiry from a researcher outside my institution
regarding analysis of silver grains in a transmission electron
micrograph.

 

I have attempted to assist this person but have not been able to obtain
the information that they are looking for although it's been interesting
trying!

 

The original bmp image sent to me by the researcher is attached
(Fig1.bmp). They want to analyse the dense black spots, which are silver
grains and find out the following;

 

How many grains/spots are there?

What size is each grain/spot?

 

They also want to define regions of interest (ROIs) and count the number
and size of the spots in specific locations within the image. You will
note that there are areas where the grains/spots are larger and in
higher density.

 

I found it impossible to segment out the spots using thresholding based
on grayscale because there are some ER fibres running behind the spots
in some places, which are of similar gray value to some of the smaller
grains/spots. I have attached a JPG image with a region drawn on it to
indicate where the problem lies. Also, some of the smaller less dense
grains are similar in intensity to these fibres.

 

I haven't managed to confirm the source of the image, e.g. film, etc.

 

All of the files that were sent to me are .bmp not TIFF.

 

So far, I have managed to count the number of grains/spots by using the
Find Maxima option. But I can't get any information on the size of each
spot using this method. I have attached some instructions showing my
approach.

 

I hope that someone will be able to provide a better solution....

 

Any assistance would be much appreciated.

 

Kind regards,

 

Jacqui

 

Jacqueline Ross

Biomedical Imaging Microscopist
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.fmhs.auckland.ac.nz/sms/biru/

[Jacqueline Ross]

Hi All,

Sorry all. I am unable to send to the listserver the PDF which
demonstrates the approach I have taken so far. If anyone is interested,
please contact me offlist.

Thanks!

Kind regards,

Jacqui

Jacqueline Ross

Biomedical Imaging Microscopist
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.fmhs.auckland.ac.nz/sms/biru/

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
Jacqui Ross
Sent: Friday, 8 April 2011 12:02 p.m.
To: [hidden email]
Subject: Segmentation and analysis of silver grains in TEM image

Dear All,

 

I recently had an enquiry from a researcher outside my institution
regarding analysis of silver grains in a transmission electron
micrograph.

 

I have attempted to assist this person but have not been able to obtain
the information that they are looking for although it's been interesting
trying!

 

The original bmp image sent to me by the researcher is attached
(Fig1.bmp). They want to analyse the dense black spots, which are silver
grains and find out the following;

 

How many grains/spots are there?

What size is each grain/spot?

 

They also want to define regions of interest (ROIs) and count the number
and size of the spots in specific locations within the image. You will
note that there are areas where the grains/spots are larger and in
higher density.

 

I found it impossible to segment out the spots using thresholding based
on grayscale because there are some ER fibres running behind the spots
in some places, which are of similar gray value to some of the smaller
grains/spots. I have attached a JPG image with a region drawn on it to
indicate where the problem lies. Also, some of the smaller less dense
grains are similar in intensity to these fibres.

 

I haven't managed to confirm the source of the image, e.g. film, etc.

 

All of the files that were sent to me are .bmp not TIFF.

 

So far, I have managed to count the number of grains/spots by using the
Find Maxima option. But I can't get any information on the size of each
spot using this method. I have attached some instructions showing my
approach.

 

I hope that someone will be able to provide a better solution....

 

Any assistance would be much appreciated.

 

Kind regards,

 

Jacqui

 

Jacqueline Ross

Biomedical Imaging Microscopist
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.fmhs.auckland.ac.nz/sms/biru/

[Klonuo]

Hi Jacqui,

I don't have answer to your question but friendly advice:

as you probably know this mailing list has many subscribers, and attaching large
files should be avoided by posting a link to the file that you have previously
uploaded elsewhere, so that those who are interested may follow your
link, and all subscribers does not have to download your attachment

Cheers

--------------------
Full-quoting makes you scroll past the same junk over and over.

[Gabriel Landini]

On Friday 08 Apr 2011 01:02:10 you wrote:
> The original bmp image sent to me by the researcher is attached
> (Fig1.bmp). They want to analyse the dense black spots, which are silver
> grains and find out the following;

This problem can be solved relatively easy via the so called Top Hat filters.
There are (in addition to the dilation/erosion based top hat filters) some
based on greyscale reconstruction. Have a look at the Morphological Image
Analysis book of P Soille.

Cheers

Gabriel

[Michael Schmid]

Hi,

just to give you one more pointer: You can get a top-hat filter by  
Process>Math>Min and then Max, both with the same radius (or other  
way around, first Max, then Min, if your foreground objects have  
lower pixel values than the background as in the sample images).
Also, in the 'Fast Filters' plugin, the 'eliminate Maxima' and  
'Eliminate Minima' filters are Top-Hat filters, and the "background  
from..." filters there are top-hat filters with additional smoothing  
of the background.

The ImageJ built-in 'Subtract background' is a more generalized type  
of a top-hat filter, better than a simple top-hat filter in areas  
with a gradient in the background.

Nevertheless, I could not get a decent removal of the fiber  
background with any of these.

The best solution I found so far is the Versatile Wand with a value  
tolerance including the full image range (255 for 8-bit) and a  
gradient tolerance of about 10.
Just click on the background and then "Make Inverse" of the  
selection. You will need some postprocessing to eliminate a few  
remaining fibers (using either their shape or their gray value).
Also looping through the maxima from "Find Maxima" in a macro and  
selecting each of them with the Versatile Wand would eliminate these  
fibers, but you have to check that the selection really stops at the  
particle center: Forget the result or decrease the threshold if the  
Wand selects too large an area. You would need a simple macro for this.

Fast Filters & Versatile Wand are available at
   http://imagejdocu.tudor.lu/doku.php?id=plugin:start

Michael
________________________________________________________________

On 8 Apr 2011, at 14:44, Gabriel Landini wrote:

> On Friday 08 Apr 2011 01:02:10 you wrote:
>> The original bmp image sent to me by the researcher is attached
>> (Fig1.bmp). They want to analyse the dense black spots, which are  
>> silver
>> grains and find out the following;
>
> This problem can be solved relatively easy via the so called Top  
> Hat filters.
> There are (in addition to the dilation/erosion based top hat  
> filters) some
> based on greyscale reconstruction. Have a look at the Morphological  
> Image
> Analysis book of P Soille.
>
> Cheers
>
> Gabriel

[Gabriel Landini]

On Friday 08 Apr 2011 14:35:46 you wrote:
> just to give you one more pointer: You can get a top-hat filter by
> Process>Math>Min and then Max, both with the same radius (or other
> way around, first Max, then Min, if your foreground objects have
> lower pixel values than the background as in the sample images).

What is described above is closing and opening in greyscale. One more step
gives you the traditional top hat: subtract from the original the result of
the above.

> Nevertheless, I could not get a decent removal of the fiber
> background with any of these.

I attached the result of:

1. Applied gaussian blur of size 1
2. Greyscale black top hat by reconstruction (see below, following Soille's
book)
3. Stretched the greyscale so the result can be seen

Step 2 is the following (and requires the Morphology collection installed)

// GreyBlackTopHatByReconstruction
// G. Landini at bham. ac. uk.
// 1/Oct/2010
setBatchMode(true);
a=getTitle();
run("Duplicate...", "title=_seed");
run("Maximum...", "radius=2");
run("Invert");
selectWindow(a);
run("Invert");
run("GreyscaleReconstruct ", "mask="+a+" seed=_seed create");
selectWindow(a);
run("Invert");
selectWindow("Reconstructed");
run("Invert");
rename("BlackTopHatReconstructed");
imageCalculator("Subtract", "BlackTopHatReconstructed",a);
setBatchMode(false);
//------

Cheers

Gabriel

[Jacqueline Ross]

Dear Gabriel, Michael and Chris,

Thanks very much to you all for your suggestions. I have run the 2
macros written by Chris and Gabriel and they both work quite well so I
will send them off to the researcher who sent me the data. It's always
interesting to see different approaches to the same problem.

I've also had a play with the Versatile Wand as per Michael's suggestion
but this isn't working quite as well as I would like, since it's a bit
fiddly to get the right selection, however I can see it will be really
useful for other tasks.

I also had a look at other Top Hat filters as suggested by Gabriel and
Michael and came across the Lipschitz filter
(http://rsbweb.nih.gov/ij/plugins/lipschitz/) that also seems to work
very well for this data with the default setting (Slope 10.00, Top Down
and Top Hat) selected.

We don't have the Soille book but I managed to find a paper by Soille,
which covers the grayscale reconstruction mentioned by Gabriel.

Thanks very much to anyone/everyone who put effort into looking into
this analysis problem. As usual, the responses have been very helpful
and I have learnt more myself in the process...!

Kind regards,

Jacqui

Jacqueline Ross

Biomedical Imaging Microscopist
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.fmhs.auckland.ac.nz/sms/biru/


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
Gabriel Landini
Sent: Saturday, 9 April 2011 2:34 a.m.
To: [hidden email]
Subject: Re: Segmentation and analysis of silver grains in TEM image

On Friday 08 Apr 2011 14:35:46 you wrote:
> just to give you one more pointer: You can get a top-hat filter by
> Process>Math>Min and then Max, both with the same radius (or other
> way around, first Max, then Min, if your foreground objects have lower

> pixel values than the background as in the sample images).

What is described above is closing and opening in greyscale. One more
step gives you the traditional top hat: subtract from the original the
result of the above.

> Nevertheless, I could not get a decent removal of the fiber background

> with any of these.

I attached the result of:

1. Applied gaussian blur of size 1
2. Greyscale black top hat by reconstruction (see below, following
Soille's
book)
3. Stretched the greyscale so the result can be seen

Step 2 is the following (and requires the Morphology collection
installed)

// GreyBlackTopHatByReconstruction
// G. Landini at bham. ac. uk.
// 1/Oct/2010
setBatchMode(true);
a=getTitle();
run("Duplicate...", "title=_seed");
run("Maximum...", "radius=2");
run("Invert");
selectWindow(a);
run("Invert");
run("GreyscaleReconstruct ", "mask="+a+" seed=_seed create");
selectWindow(a); run("Invert"); selectWindow("Reconstructed");
run("Invert"); rename("BlackTopHatReconstructed");
imageCalculator("Subtract", "BlackTopHatReconstructed",a);
setBatchMode(false);
//------

Cheers

Gabriel