identifying particles, statistical analysis, averaging topographic views, etc

[luisalles]

Hello,
I have spent now a few days playing around with Fiji on images from atomic
force microscopy (AFM) collected on a protein of interest. I am able to take
profit from most basic functions, but I realize that I am far from being
capable of doing what I would need to. My impression is that I would
probably need to invest a very long time to get trained, something that
unfortunately I cannot afford if I wish to progress in other aspects of my
bench work. That is why I have opted by looking for help here.
Please let me explain one case:
One of our proteins of interest is assembling on surfaces to give rise to
motifs like those of the next figures:
<http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
Individual protein units/blocks are supposed to be more or less globular,
like beads, about 7-8 nm wide and high. As the image indicates, these little
beads can assemble together to give rise to different patches. Sometimes
beads arrange linearly, as "pearl necklaces" that possibly stack on each
other, producing small platforms composed by variable number of beads.
However, in many instances, especially when the protein load is smaller,
holes are observed (see next figure). These arrangements might be made of 5
or 6 beads disposed like forming a big pentagon or hexagon, with internal
central voids.
<http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
What information would be pursued? Obviously, the most information we could
extract from the images the best. But to be more precise, it would be great
if one could delimit/contour single individual beads using an adapted
approach like those used for single particle analysis. That would permit to
carry out statistical analysis of type and size of patches formed (be aware
that there are other images obtained in mixtures with other potential
protein partners, which in some cases modify assembly patterns). But
probably most important would be if we could generate averaged topographic
images of individual beads and (let us say) an hexagonal patch with a
central hole. I suppose that might be feasible by averaging over data for
individual beads/hexagons extracted from the same image, or from different
images if necessary, after having minimized the topographic matching of
individual hits (by applying rotation/translation operations when necessary
before averaging).
Could any of you help us in carrying out this type of analysis? Obviously
you would be considered as collaborator, meaning that you would appear as
coauthor of scientific publications presenting such data.
Thank you very much in advance,



--
Sent from: http://imagej.1557.x6.nabble.com/

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

[Michael Schmid]

Hi anonymous,

you could try the Feature Finder plugin (template matching), it can at
least find the 'globular' objects that can be easily distinguished, It
can also create an average of all these objects, see inset in the
attachment.
You might also select a globular object next to a void and check for
equivalents.

As your data are quite noisy, you may want to go beyond the slider value
with the tolerance (up to 200 or so). It also helps to create an average
and use the average as prototype (maybe increase the contrast of the
prototype by multiplying with 1.2 or so).


http://imagejdocu.tudor.lu/doku.php?id=plugin:analysis:feature_finder:start

Note that 'Compile&Run' might not work on Fiji; it should work with
plain ImageJ.

Michael
________________________________________________________________
On 09/01/2018 13:08, luisalles wrote:

> Hello,
> I have spent now a few days playing around with Fiji on images from atomic
> force microscopy (AFM) collected on a protein of interest. I am able to take
> profit from most basic functions, but I realize that I am far from being
> capable of doing what I would need to. My impression is that I would
> probably need to invest a very long time to get trained, something that
> unfortunately I cannot afford if I wish to progress in other aspects of my
> bench work. That is why I have opted by looking for help here.
> Please let me explain one case:
> One of our proteins of interest is assembling on surfaces to give rise to
> motifs like those of the next figures:
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
> Individual protein units/blocks are supposed to be more or less globular,
> like beads, about 7-8 nm wide and high. As the image indicates, these little
> beads can assemble together to give rise to different patches. Sometimes
> beads arrange linearly, as "pearl necklaces" that possibly stack on each
> other, producing small platforms composed by variable number of beads.
> However, in many instances, especially when the protein load is smaller,
> holes are observed (see next figure). These arrangements might be made of 5
> or 6 beads disposed like forming a big pentagon or hexagon, with internal
> central voids.
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
> What information would be pursued? Obviously, the most information we could
> extract from the images the best. But to be more precise, it would be great
> if one could delimit/contour single individual beads using an adapted
> approach like those used for single particle analysis. That would permit to
> carry out statistical analysis of type and size of patches formed (be aware
> that there are other images obtained in mixtures with other potential
> protein partners, which in some cases modify assembly patterns). But
> probably most important would be if we could generate averaged topographic
> images of individual beads and (let us say) an hexagonal patch with a
> central hole. I suppose that might be feasible by averaging over data for
> individual beads/hexagons extracted from the same image, or from different
> images if necessary, after having minimized the topographic matching of
> individual hits (by applying rotation/translation operations when necessary
> before averaging).
> Could any of you help us in carrying out this type of analysis? Obviously
> you would be considered as collaborator, meaning that you would appear as
> coauthor of scientific publications presenting such data.
> Thank you very much in advance,
>
>
>
> --
> Sent from: http://imagej.1557.x6.nabble.com/
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

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

[CARL Philippe (LBP)]

Hi Michael,
How are you doing? Happy New Year!!!
Since Luis is a compatriot, I just answered him in french that ImageJ may not be the best tool to analyze its AFM pictures.
In fact there is a dedicated SPM data visualization and analysis open source software called Gwyddion (http://gwyddion.net/).
Among other things, Gwyddion will able him to apply some specific filtering to get rid of the horizontal stripes that can be seen within his pictures and which are specific to the acquisition technique.
My best regards,
Philippe

Philippe CARL
Laboratoire de Bioimagerie et Pathologies
UMR 7021 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 Michael Schmid
Envoyé : mardi 9 janvier 2018 15:34
À : [hidden email]
Objet : Re: identifying particles, statistical analysis, averaging topographic views, etc

Hi anonymous,

you could try the Feature Finder plugin (template matching), it can at least find the 'globular' objects that can be easily distinguished, It can also create an average of all these objects, see inset in the attachment.
You might also select a globular object next to a void and check for equivalents.

As your data are quite noisy, you may want to go beyond the slider value with the tolerance (up to 200 or so). It also helps to create an average and use the average as prototype (maybe increase the contrast of the prototype by multiplying with 1.2 or so).


http://imagejdocu.tudor.lu/doku.php?id=plugin:analysis:feature_finder:start

Note that 'Compile&Run' might not work on Fiji; it should work with plain ImageJ.

Michael
________________________________________________________________
On 09/01/2018 13:08, luisalles wrote:

> Hello,
> I have spent now a few days playing around with Fiji on images from
> atomic force microscopy (AFM) collected on a protein of interest. I am
> able to take profit from most basic functions, but I realize that I am
> far from being capable of doing what I would need to. My impression is
> that I would probably need to invest a very long time to get trained,
> something that unfortunately I cannot afford if I wish to progress in
> other aspects of my bench work. That is why I have opted by looking for help here.
> Please let me explain one case:
> One of our proteins of interest is assembling on surfaces to give rise
> to motifs like those of the next figures:
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
> Individual protein units/blocks are supposed to be more or less
> globular, like beads, about 7-8 nm wide and high. As the image
> indicates, these little beads can assemble together to give rise to
> different patches. Sometimes beads arrange linearly, as "pearl
> necklaces" that possibly stack on each other, producing small platforms composed by variable number of beads.
> However, in many instances, especially when the protein load is
> smaller, holes are observed (see next figure). These arrangements
> might be made of 5 or 6 beads disposed like forming a big pentagon or
> hexagon, with internal central voids.
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
> What information would be pursued? Obviously, the most information we
> could extract from the images the best. But to be more precise, it
> would be great if one could delimit/contour single individual beads
> using an adapted approach like those used for single particle
> analysis. That would permit to carry out statistical analysis of type
> and size of patches formed (be aware that there are other images
> obtained in mixtures with other potential protein partners, which in
> some cases modify assembly patterns). But probably most important
> would be if we could generate averaged topographic images of
> individual beads and (let us say) an hexagonal patch with a central
> hole. I suppose that might be feasible by averaging over data for
> individual beads/hexagons extracted from the same image, or from
> different images if necessary, after having minimized the topographic
> matching of individual hits (by applying rotation/translation operations when necessary before averaging).
> Could any of you help us in carrying out this type of analysis?
> Obviously you would be considered as collaborator, meaning that you
> would appear as coauthor of scientific publications presenting such data.
> Thank you very much in advance,
>
>
>
> --
> Sent from: http://imagej.1557.x6.nabble.com/
>
> --
> 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

[Michael Schmid]

Hi Philippe, Luis,

yes, Gwyddion is nice, but ImageJ has far more possibilities for image
analysis (I use ImageJ a lot for STM images).

For removing the horizontal stripes, it is easy to write a plugin that
subtracts the average of the lines (or even better, subtracts the median
of the lines). Also the FFT filtering of ImageJ helps a lot (there is
also a 'suppress stripes' in the FFT>Bandpass filter)...

Best,

Michael
________________________________________________________________
On 09/01/2018 16:10, Philippe CARL wrote:

> Hi Michael,
> How are you doing? Happy New Year!!!
> Since Luis is a compatriot, I just answered him in french that ImageJ may not be the best tool to analyze its AFM pictures.
> In fact there is a dedicated SPM data visualization and analysis open source software called Gwyddion (http://gwyddion.net/).
> Among other things, Gwyddion will able him to apply some specific filtering to get rid of the horizontal stripes that can be seen within his pictures and which are specific to the acquisition technique.
> My best regards,
> Philippe
>
> Philippe CARL
> Laboratoire de Bioimagerie et Pathologies
> UMR 7021 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 Michael Schmid
> Envoyé : mardi 9 janvier 2018 15:34
> À : [hidden email]
> Objet : Re: identifying particles, statistical analysis, averaging topographic views, etc
>
> Hi anonymous,
>
> you could try the Feature Finder plugin (template matching), it can at least find the 'globular' objects that can be easily distinguished, It can also create an average of all these objects, see inset in the attachment.
> You might also select a globular object next to a void and check for equivalents.
>
> As your data are quite noisy, you may want to go beyond the slider value with the tolerance (up to 200 or so). It also helps to create an average and use the average as prototype (maybe increase the contrast of the prototype by multiplying with 1.2 or so).
>
>
> http://imagejdocu.tudor.lu/doku.php?id=plugin:analysis:feature_finder:start
>
> Note that 'Compile&Run' might not work on Fiji; it should work with plain ImageJ.
>
> Michael
> ________________________________________________________________
> On 09/01/2018 13:08, luisalles wrote:
>> Hello,
>> I have spent now a few days playing around with Fiji on images from
>> atomic force microscopy (AFM) collected on a protein of interest. I am
>> able to take profit from most basic functions, but I realize that I am
>> far from being capable of doing what I would need to. My impression is
>> that I would probably need to invest a very long time to get trained,
>> something that unfortunately I cannot afford if I wish to progress in
>> other aspects of my bench work. That is why I have opted by looking for help here.
>> Please let me explain one case:
>> One of our proteins of interest is assembling on surfaces to give rise
>> to motifs like those of the next figures:
>> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
>> Individual protein units/blocks are supposed to be more or less
>> globular, like beads, about 7-8 nm wide and high. As the image
>> indicates, these little beads can assemble together to give rise to
>> different patches. Sometimes beads arrange linearly, as "pearl
>> necklaces" that possibly stack on each other, producing small platforms composed by variable number of beads.
>> However, in many instances, especially when the protein load is
>> smaller, holes are observed (see next figure). These arrangements
>> might be made of 5 or 6 beads disposed like forming a big pentagon or
>> hexagon, with internal central voids.
>> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
>> What information would be pursued? Obviously, the most information we
>> could extract from the images the best. But to be more precise, it
>> would be great if one could delimit/contour single individual beads
>> using an adapted approach like those used for single particle
>> analysis. That would permit to carry out statistical analysis of type
>> and size of patches formed (be aware that there are other images
>> obtained in mixtures with other potential protein partners, which in
>> some cases modify assembly patterns). But probably most important
>> would be if we could generate averaged topographic images of
>> individual beads and (let us say) an hexagonal patch with a central
>> hole. I suppose that might be feasible by averaging over data for
>> individual beads/hexagons extracted from the same image, or from
>> different images if necessary, after having minimized the topographic
>> matching of individual hits (by applying rotation/translation operations when necessary before averaging).
>> Could any of you help us in carrying out this type of analysis?
>> Obviously you would be considered as collaborator, meaning that you
>> would appear as coauthor of scientific publications presenting such data.
>> Thank you very much in advance,
>>
>>
>>
>> --
>> Sent from: http://imagej.1557.x6.nabble.com/
>>
>> --
>> 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
>

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

[Kenneth Sloan-2]

This is an interesting problem.  The solution depends critically on what SCALE information you really need.
It's not at all clear to me that you *need* to delineate (or even identify) the individual protein blobs.  If they are as regular in size and shape as you imply, then it might be sufficient to measure the area covered by protein blobs (and the shape and areas of the holes).

Since you have height information, you could probably assess stacking.

So...my *first* approach would be to characterize pixels in the image according to "max height within 5mm".  I would then try to delineate the boundaries between protein globule-dominated area and the holes.  I think I would first try to create a cartoon of your image, with regions created by thresholding on "max nearby height" at levels corresponding to 0, 1, 2, 3 high stacks of globules (I think I only see 0, 1, and 2 high regions in your sample images).

This would give you topographic information on the holes, and decent estimates on the number of globules.

My next step would be to assume hexagonal packing of the globules.  I would look for "obvious" globules, and try to use these as seeds to suggest nearby globules.  "Verification vision" is often easier than bottom up feature recognition.  Once you have a critical mass of well-identified globules, you can make strong predictions (which you can then verify) about the locations of adjacent globules.  This mostly works only for 1-high regions.  Once the globules become stacked, this strategy fails.

All of the above assumes:

a) you have the capability in-house to write some custom Java code
b) I have correctly characterized the kind of information you care about.

You have a lot of "domain knowledge" about these images - the standard ImageJ tools necessarily try to do image analysis withOUT such knowledge.  As others have demonstrated, it's possible to manually manipulate the standard ImageJ tools to get reasonable results.  But, in my opinion, this is trying to solve the problem with one hand tied behind your back.

--
Kenneth Sloan
[hidden email]
Vision is the art of seeing what is invisible to others.

> On 9 Jan 2018, at 06:08 , luisalles <[hidden email]> wrote:
>
> Hello,
> I have spent now a few days playing around with Fiji on images from atomic
> force microscopy (AFM) collected on a protein of interest. I am able to take
> profit from most basic functions, but I realize that I am far from being
> capable of doing what I would need to. My impression is that I would
> probably need to invest a very long time to get trained, something that
> unfortunately I cannot afford if I wish to progress in other aspects of my
> bench work. That is why I have opted by looking for help here.
> Please let me explain one case:
> One of our proteins of interest is assembling on surfaces to give rise to
> motifs like those of the next figures:
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
> Individual protein units/blocks are supposed to be more or less globular,
> like beads, about 7-8 nm wide and high. As the image indicates, these little
> beads can assemble together to give rise to different patches. Sometimes
> beads arrange linearly, as "pearl necklaces" that possibly stack on each
> other, producing small platforms composed by variable number of beads.
> However, in many instances, especially when the protein load is smaller,
> holes are observed (see next figure). These arrangements might be made of 5
> or 6 beads disposed like forming a big pentagon or hexagon, with internal
> central voids.
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
> What information would be pursued? Obviously, the most information we could
> extract from the images the best. But to be more precise, it would be great
> if one could delimit/contour single individual beads using an adapted
> approach like those used for single particle analysis. That would permit to
> carry out statistical analysis of type and size of patches formed (be aware
> that there are other images obtained in mixtures with other potential
> protein partners, which in some cases modify assembly patterns). But
> probably most important would be if we could generate averaged topographic
> images of individual beads and (let us say) an hexagonal patch with a
> central hole. I suppose that might be feasible by averaging over data for
> individual beads/hexagons extracted from the same image, or from different
> images if necessary, after having minimized the topographic matching of
> individual hits (by applying rotation/translation operations when necessary
> before averaging).
> Could any of you help us in carrying out this type of analysis? Obviously
> you would be considered as collaborator, meaning that you would appear as
> coauthor of scientific publications presenting such data.
> Thank you very much in advance,
>
>
>
> --
> Sent from: http://imagej.1557.x6.nabble.com/
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Krs5]

Dear Luis,

It seems that also the ITCN plugin does not do a bad job in identifying your structures.

Best wishes

Kees


Dr Ir K.R. Straatman
Senior Experimental Officer
Advanced Imaging Facility
Centre for Core Biotechnology Services
University of Leicester
www.le.ac.uk/advanced-imaging-facility


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Philippe CARL
Sent: 09 January 2018 15:10
To: [hidden email]
Subject: Re: identifying particles, statistical analysis, averaging topographic views, etc

Hi Michael,
How are you doing? Happy New Year!!!
Since Luis is a compatriot, I just answered him in french that ImageJ may not be the best tool to analyze its AFM pictures.
In fact there is a dedicated SPM data visualization and analysis open source software called Gwyddion (http://gwyddion.net/).
Among other things, Gwyddion will able him to apply some specific filtering to get rid of the horizontal stripes that can be seen within his pictures and which are specific to the acquisition technique.
My best regards,
Philippe

Philippe CARL
Laboratoire de Bioimagerie et Pathologies UMR 7021 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 Michael Schmid Envoyé : mardi 9 janvier 2018 15:34 À : [hidden email] Objet : Re: identifying particles, statistical analysis, averaging topographic views, etc

Hi anonymous,

you could try the Feature Finder plugin (template matching), it can at least find the 'globular' objects that can be easily distinguished, It can also create an average of all these objects, see inset in the attachment.
You might also select a globular object next to a void and check for equivalents.

As your data are quite noisy, you may want to go beyond the slider value with the tolerance (up to 200 or so). It also helps to create an average and use the average as prototype (maybe increase the contrast of the prototype by multiplying with 1.2 or so).


http://imagejdocu.tudor.lu/doku.php?id=plugin:analysis:feature_finder:start

Note that 'Compile&Run' might not work on Fiji; it should work with plain ImageJ.

Michael
________________________________________________________________
On 09/01/2018 13:08, luisalles wrote:

> Hello,
> I have spent now a few days playing around with Fiji on images from
> atomic force microscopy (AFM) collected on a protein of interest. I am
> able to take profit from most basic functions, but I realize that I am
> far from being capable of doing what I would need to. My impression is
> that I would probably need to invest a very long time to get trained,
> something that unfortunately I cannot afford if I wish to progress in
> other aspects of my bench work. That is why I have opted by looking for help here.
> Please let me explain one case:
> One of our proteins of interest is assembling on surfaces to give rise
> to motifs like those of the next figures:
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_020.jpg>
> Individual protein units/blocks are supposed to be more or less
> globular, like beads, about 7-8 nm wide and high. As the image
> indicates, these little beads can assemble together to give rise to
> different patches. Sometimes beads arrange linearly, as "pearl
> necklaces" that possibly stack on each other, producing small platforms composed by variable number of beads.
> However, in many instances, especially when the protein load is
> smaller, holes are observed (see next figure). These arrangements
> might be made of 5 or 6 beads disposed like forming a big pentagon or
> hexagon, with internal central voids.
> <http://imagej.1557.x6.nabble.com/file/t381883/Pd190417_023.jpg>
> What information would be pursued? Obviously, the most information we
> could extract from the images the best. But to be more precise, it
> would be great if one could delimit/contour single individual beads
> using an adapted approach like those used for single particle
> analysis. That would permit to carry out statistical analysis of type
> and size of patches formed (be aware that there are other images
> obtained in mixtures with other potential protein partners, which in
> some cases modify assembly patterns). But probably most important
> would be if we could generate averaged topographic images of
> individual beads and (let us say) an hexagonal patch with a central
> hole. I suppose that might be feasible by averaging over data for
> individual beads/hexagons extracted from the same image, or from
> different images if necessary, after having minimized the topographic
> matching of individual hits (by applying rotation/translation operations when necessary before averaging).
> Could any of you help us in carrying out this type of analysis?
> Obviously you would be considered as collaborator, meaning that you
> would appear as coauthor of scientific publications presenting such data.
> Thank you very much in advance,
>
>
>
> --
> Sent from: http://imagej.1557.x6.nabble.com/
>
> --
> 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

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

[luisalles]

Dear people,
First, thanks a lot for your generous feedback !
I will try to apply the different approaches suggested. However, I already
realized my case is not trivial...
Michael suggests to try to use a "feature finder" tool. I will follow his
advice, but I must say that I tried already to play with cvMatch_template
without much success. In fact, it seems to find properly the furnished
template, and depending on the tolerance it goes beyond and finds also some
sucessful hits. However, it also points to false hits, and I have the
impression is not considering the possibility of "rotated" objects. It only
points to motifs with same orientation. If that were true, it would mean
that I'd need to build templates for each possible orientation and type of
"assembly". This problem of orientation also hampered other approach that I
attempted with the intention to obtain a view averaged over many similar
objects: 1) extracting from an image individual files for cropped single
"hexagons with holes";2) mounting a stack of 30 such images and 3) matching
the images with "Align slices in stack" plugin; 4) combining the slices
using Z-project. Overall it worked, however it was clear that the averaged
view was much worse than possible, due to this "image-rotation" problem.
Concerning Philippe's advice of turning myself towards Gwyddion, it will
depend on how things progress over next days with ImageJ. In fact, ImageJ
seems extremely powerful, and it adds that one can count on the presence of
this "supporting comunity".
In the end, after having tried for a couple of days, I admit Kenneth's
advice of trying to analyze the images in a semi-automatic manner could be
the best. I will try to go image by image, trying to delimitate "islands or
blobs" and simply estimating numbers of particles inside.  I did not
understand well what he meant when writting "max height within 5 mm", but I
suppose he refers to trying to make views at different thresholds to somehow
"cut" the view at different heights?
Finaly, I installed the ITCN plugin, as suggested by Kees. It worked pretty
well in counting spheres on the first image, which is almost fully covered
by arrangements. This is already something I could exploit. Unfortunately,
it only worked without ROI selected on the image, when indicating only the
approximate size of "cells". When a circular ROI was proposed around a
single globule, it did not work, it only poped out the original position. I
don't know whether a kind of tolerance needs to be modified somewhere in the
plugin code to allow this to work. ITCN did not work neither with the second
image, regardless of whether a ROI was proposed or not. In fact, most of
"cell hits" lied on black background regions.
By the way, I've also tried to play with the "trainable Weka segmentation".
I created different type of traces for each of two or even three training
classes. No problem to classify separatedly the background regions protein
globules, but I could not make it work in recognizing grooves left in
between globules.
I will continue to "experience" with this software, but my lesson up to now
is that I might waste longer my time in trying to do automatic things
instead of going image by image and doing semi-manually.
Well, do not hesitate to propose any other suggestion you might figure out.
Once again, thank you very very much for your help,



--
Sent from: http://imagej.1557.x6.nabble.com/

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

[Krs5]

Concerning the ITCN plugin, if you look carefully at your second image (or adjust Brightness/Contrast) you can see that there are structures in the background and they are correctly identified by the plugin. So the trick will be to get rid of this background signal.

However,  there is a later version of the ITCN plugin, although it is not available online anymore. The website disappeared a few years ago. This plugin includes a threshold option and setting this at 2 results in the attached image. The mailing list does not accept the plugin as attachment so I will try to send it directly to you. You probably can further improve this result by filtering  (FFT Bandpass Filter?)/background subtraction.

The ROI is only if you don't want to analyse the whole image but just a part of it. It is not to identify particles of interest.

Best wishes

Kees

-----Original Message-----
From: luisalles [mailto:[hidden email]]
Sent: 10 January 2018 11:41
To: [hidden email]
Subject: Re: identifying particles, statistical analysis, averaging topographic views, etc

Dear people,
First, thanks a lot for your generous feedback !
I will try to apply the different approaches suggested. However, I already realized my case is not trivial...
Michael suggests to try to use a "feature finder" tool. I will follow his advice, but I must say that I tried already to play with cvMatch_template without much success. In fact, it seems to find properly the furnished template, and depending on the tolerance it goes beyond and finds also some sucessful hits. However, it also points to false hits, and I have the impression is not considering the possibility of "rotated" objects. It only points to motifs with same orientation. If that were true, it would mean that I'd need to build templates for each possible orientation and type of "assembly". This problem of orientation also hampered other approach that I attempted with the intention to obtain a view averaged over many similar
objects: 1) extracting from an image individual files for cropped single "hexagons with holes";2) mounting a stack of 30 such images and 3) matching the images with "Align slices in stack" plugin; 4) combining the slices using Z-project. Overall it worked, however it was clear that the averaged view was much worse than possible, due to this "image-rotation" problem.
Concerning Philippe's advice of turning myself towards Gwyddion, it will depend on how things progress over next days with ImageJ. In fact, ImageJ seems extremely powerful, and it adds that one can count on the presence of this "supporting comunity".
In the end, after having tried for a couple of days, I admit Kenneth's advice of trying to analyze the images in a semi-automatic manner could be the best. I will try to go image by image, trying to delimitate "islands or blobs" and simply estimating numbers of particles inside.  I did not understand well what he meant when writting "max height within 5 mm", but I suppose he refers to trying to make views at different thresholds to somehow "cut" the view at different heights?
Finaly, I installed the ITCN plugin, as suggested by Kees. It worked pretty well in counting spheres on the first image, which is almost fully covered by arrangements. This is already something I could exploit. Unfortunately, it only worked without ROI selected on the image, when indicating only the approximate size of "cells". When a circular ROI was proposed around a single globule, it did not work, it only poped out the original position. I don't know whether a kind of tolerance needs to be modified somewhere in the plugin code to allow this to work. ITCN did not work neither with the second image, regardless of whether a ROI was proposed or not. In fact, most of "cell hits" lied on black background regions.
By the way, I've also tried to play with the "trainable Weka segmentation".
I created different type of traces for each of two or even three training classes. No problem to classify separatedly the background regions protein globules, but I could not make it work in recognizing grooves left in between globules.
I will continue to "experience" with this software, but my lesson up to now is that I might waste longer my time in trying to do automatic things instead of going image by image and doing semi-manually.
Well, do not hesitate to propose any other suggestion you might figure out.
Once again, thank you very very much for your help,



--
Sent from: http://imagej.1557.x6.nabble.com/

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

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

[Krs5]

Sorry, I was wrong. The later version of ITCN plugin is still online and can be downloaded from http://bioimage.ucsb.edu/docs/automatic-nuclei-counter-plugin-imagej.

Best wishes

Kees



-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Straatman, Kees (Dr.)
Sent: 10 January 2018 17:22
To: [hidden email]
Subject: Re: identifying particles, statistical analysis, averaging topographic views, etc

Concerning the ITCN plugin, if you look carefully at your second image (or adjust Brightness/Contrast) you can see that there are structures in the background and they are correctly identified by the plugin. So the trick will be to get rid of this background signal.

However,  there is a later version of the ITCN plugin, although it is not available online anymore. The website disappeared a few years ago. This plugin includes a threshold option and setting this at 2 results in the attached image. The mailing list does not accept the plugin as attachment so I will try to send it directly to you. You probably can further improve this result by filtering  (FFT Bandpass Filter?)/background subtraction.

The ROI is only if you don't want to analyse the whole image but just a part of it. It is not to identify particles of interest.

Best wishes

Kees

-----Original Message-----
From: luisalles [mailto:[hidden email]]
Sent: 10 January 2018 11:41
To: [hidden email]
Subject: Re: identifying particles, statistical analysis, averaging topographic views, etc

Dear people,
First, thanks a lot for your generous feedback !
I will try to apply the different approaches suggested. However, I already realized my case is not trivial...
Michael suggests to try to use a "feature finder" tool. I will follow his advice, but I must say that I tried already to play with cvMatch_template without much success. In fact, it seems to find properly the furnished template, and depending on the tolerance it goes beyond and finds also some sucessful hits. However, it also points to false hits, and I have the impression is not considering the possibility of "rotated" objects. It only points to motifs with same orientation. If that were true, it would mean that I'd need to build templates for each possible orientation and type of "assembly". This problem of orientation also hampered other approach that I attempted with the intention to obtain a view averaged over many similar
objects: 1) extracting from an image individual files for cropped single "hexagons with holes";2) mounting a stack of 30 such images and 3) matching the images with "Align slices in stack" plugin; 4) combining the slices using Z-project. Overall it worked, however it was clear that the averaged view was much worse than possible, due to this "image-rotation" problem.
Concerning Philippe's advice of turning myself towards Gwyddion, it will depend on how things progress over next days with ImageJ. In fact, ImageJ seems extremely powerful, and it adds that one can count on the presence of this "supporting comunity".
In the end, after having tried for a couple of days, I admit Kenneth's advice of trying to analyze the images in a semi-automatic manner could be the best. I will try to go image by image, trying to delimitate "islands or blobs" and simply estimating numbers of particles inside.  I did not understand well what he meant when writting "max height within 5 mm", but I suppose he refers to trying to make views at different thresholds to somehow "cut" the view at different heights?
Finaly, I installed the ITCN plugin, as suggested by Kees. It worked pretty well in counting spheres on the first image, which is almost fully covered by arrangements. This is already something I could exploit. Unfortunately, it only worked without ROI selected on the image, when indicating only the approximate size of "cells". When a circular ROI was proposed around a single globule, it did not work, it only poped out the original position. I don't know whether a kind of tolerance needs to be modified somewhere in the plugin code to allow this to work. ITCN did not work neither with the second image, regardless of whether a ROI was proposed or not. In fact, most of "cell hits" lied on black background regions.
By the way, I've also tried to play with the "trainable Weka segmentation".
I created different type of traces for each of two or even three training classes. No problem to classify separatedly the background regions protein globules, but I could not make it work in recognizing grooves left in between globules.
I will continue to "experience" with this software, but my lesson up to now is that I might waste longer my time in trying to do automatic things instead of going image by image and doing semi-manually.
Well, do not hesitate to propose any other suggestion you might figure out.
Once again, thank you very very much for your help,



--
Sent from: http://imagej.1557.x6.nabble.com/

--
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