soma area detection

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|

soma area detection

Qing Liu
Hi,

As a new user of ImageJ. I am trying to automate some of my measurements
about neuronal sizes. Has anyone tried before or know how to identify cell
soma apart from their processes? The regular threshold tool or edge detector
would fail because of the presence of the process or the nuclei. Thanks a
lot for your time.

Qing
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Nathaniel Ryckman
Qing Liu wrote
Hi,

As a new user of ImageJ. I am trying to automate some of my measurements
about neuronal sizes. Has anyone tried before or know how to identify cell
soma apart from their processes? The regular threshold tool or edge detector
would fail because of the presence of the process or the nuclei. Thanks a
lot for your time.

Qing
Here's a quick idea. I don't know how well it will work, but I'm thinking that the Watershed program might cut off the the processes. Then you could iterate through the ROI's created to see which one is most circular (soma).

If you give me a picture, I'll come up with something.
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Nathaniel Ryckman
In reply to this post by Qing Liu
Qing Liu wrote
Hi,

As a new user of ImageJ. I am trying to automate some of my measurements
about neuronal sizes. Has anyone tried before or know how to identify cell
soma apart from their processes? The regular threshold tool or edge detector
would fail because of the presence of the process or the nuclei. Thanks a
lot for your time.

Qing
Here's a quick idea. I don't know how well it will work, but I'm thinking that the Watershed program might cut off the the processes. Then you could iterate through the ROI's created to see which one is most circular (soma).

If you give me a picture, I'll come up with something.
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Qing Liu
In reply to this post by Qing Liu
Dear all, I have found a solution to select neuronal soma but need your
suggestions to do so in imageJ.

http://dl.dropbox.com/u/266880/Soma.pdf

In the above paper, they mentioned

the images of the neurons stained
> for beta-III tubulin are first Gaussian filtered to suppress image
> noise. They are background corrected using a top hat morphological filter
> with a disk structuring element larger than the
> size of the largest cell body. A morphological opening
> with a structuring element of a diameter smaller than the cell
> body but larger than all neurites permits suppressing neuritelike
> structures.
>

I am a little confused about the second and third step and wonder if anyone
can help.

Thank you.

Qing
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Michael Schmid
Hi Quing,

here is an attempt to translate the recipe into ImageJ commands:

Top-hat with circular kernel of radius r:
- duplicate the image
   Assuming that your foreground objects have higher pixel value
   than the background:
- Process>Filters>Minimum on the copy, radius r, followed by
- Process>Filters>Maximum on the copy, same radius r
- Image calculator to subtract the copy from the original
If your foreground pixel values are smaller than those of the  
background, exchange Min&Max, and subtract the original from the copy.

Morphological opening:
- Process>Filters>Minimum with radius r1 (smaller than r for the top-
hat), then
- Process>Filters>Maximum with the same radius r1


Michael
________________________________________________________________

On 19 Apr 2011, at 00:39, Qing Liu wrote:

> Dear all, I have found a solution to select neuronal soma but need  
> your
> suggestions to do so in imageJ.
>
> http://dl.dropbox.com/u/266880/Soma.pdf
>
> In the above paper, they mentioned
>
> the images of the neurons stained
>> for beta-III tubulin are first Gaussian filtered to suppress image
>> noise. They are background corrected using a top hat morphological  
>> filter
>> with a disk structuring element larger than the
>> size of the largest cell body. A morphological opening
>> with a structuring element of a diameter smaller than the cell
>> body but larger than all neurites permits suppressing neuritelike
>> structures.
>>
>
> I am a little confused about the second and third step and wonder  
> if anyone
> can help.
>
> Thank you.
>
> Qing
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Qing Liu
Hi, Michael,

Thank you for your suggestion. I am able to achieve better results with
Matlab morphological opening functions (strel and imopen):
http://www.mathworks.com/help/toolbox/images/ref/imopen.html.

With brightness/contrast/level adjustment in ImageJ and the morphological
opening settings in Matlab, I could suppress the neurites but the nuclei
seem to be problematic for applying threshold segmentation tool to measure
the soma size. Any suggestions for filling the holes/nuclei in the grayscale
image before the reconstruction? Thanks in advance.

Before: http://dl.dropbox.com/u/266880/start.tif
After: http://dl.dropbox.com/u/266880/MorphologicalOpening.tif

Qing


On Tue, Apr 19, 2011 at 4:39 AM, Michael Schmid <[hidden email]>wrote:

> Hi Quing,
>
> here is an attempt to translate the recipe into ImageJ commands:
>
> Top-hat with circular kernel of radius r:
> - duplicate the image
>  Assuming that your foreground objects have higher pixel value
>  than the background:
> - Process>Filters>Minimum on the copy, radius r, followed by
> - Process>Filters>Maximum on the copy, same radius r
> - Image calculator to subtract the copy from the original
> If your foreground pixel values are smaller than those of the background,
> exchange Min&Max, and subtract the original from the copy.
>
> Morphological opening:
> - Process>Filters>Minimum with radius r1 (smaller than r for the top-hat),
> then
> - Process>Filters>Maximum with the same radius r1
>
>
> Michael
> ________________________________________________________________
>
>
> On 19 Apr 2011, at 00:39, Qing Liu wrote:
>
>  Dear all, I have found a solution to select neuronal soma but need your
>> suggestions to do so in imageJ.
>>
>> http://dl.dropbox.com/u/266880/Soma.pdf
>>
>> In the above paper, they mentioned
>>
>> the images of the neurons stained
>>
>>> for beta-III tubulin are first Gaussian filtered to suppress image
>>> noise. They are background corrected using a top hat morphological filter
>>> with a disk structuring element larger than the
>>> size of the largest cell body. A morphological opening
>>> with a structuring element of a diameter smaller than the cell
>>> body but larger than all neurites permits suppressing neuritelike
>>> structures.
>>>
>>>
>> I am a little confused about the second and third step and wonder if
>> anyone
>> can help.
>>
>> Thank you.
>>
>> Qing
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: soma area detection

Michael Schmid
Hi Quing,

As I am not a biologist I don't know what structures you want to  
extract from your image - I hope for the biologists on the list to  
solve that problem.

Concerning ImageJ vs. Matlab:
As far as I can say from the sample image on the MatLab site, the  
operation is exactly the same as minimum+maximum in ImageJ, but there  
is a slight difference of kernel size: Matlab says the radius is 5; I  
have to use a radius of 4.5 in ImageJ for the same result. This is  
not astonishing - there are different conventions for the radius.  
Otherwise, only the brightness&contrast of the output image has been  
readjusted to fit the 0-255 range of 8-bit data, then you get the  
same as on the Matlab site.

Michael
________________________________________________________________

On 19 Apr 2011, at 23:53, Qing Liu wrote:

> Hi, Michael,
>
> Thank you for your suggestion. I am able to achieve better results  
> with
> Matlab morphological opening functions (strel and imopen):
> http://www.mathworks.com/help/toolbox/images/ref/imopen.html.
>
> With brightness/contrast/level adjustment in ImageJ and the  
> morphological
> opening settings in Matlab, I could suppress the neurites but the  
> nuclei
> seem to be problematic for applying threshold segmentation tool to  
> measure
> the soma size. Any suggestions for filling the holes/nuclei in the  
> grayscale
> image before the reconstruction? Thanks in advance.
>
> Before: http://dl.dropbox.com/u/266880/start.tif
> After: http://dl.dropbox.com/u/266880/MorphologicalOpening.tif
>
> Qing
>
>
> On Tue, Apr 19, 2011 at 4:39 AM, Michael Schmid  
> <[hidden email]>wrote:
>
>> Hi Quing,
>>
>> here is an attempt to translate the recipe into ImageJ commands:
>>
>> Top-hat with circular kernel of radius r:
>> - duplicate the image
>>  Assuming that your foreground objects have higher pixel value
>>  than the background:
>> - Process>Filters>Minimum on the copy, radius r, followed by
>> - Process>Filters>Maximum on the copy, same radius r
>> - Image calculator to subtract the copy from the original
>> If your foreground pixel values are smaller than those of the  
>> background,
>> exchange Min&Max, and subtract the original from the copy.
>>
>> Morphological opening:
>> - Process>Filters>Minimum with radius r1 (smaller than r for the  
>> top-hat),
>> then
>> - Process>Filters>Maximum with the same radius r1
>>
>>
>> Michael
>> ________________________________________________________________
>>
>>
>> On 19 Apr 2011, at 00:39, Qing Liu wrote:
>>
>>  Dear all, I have found a solution to select neuronal soma but  
>> need your
>>> suggestions to do so in imageJ.
>>>
>>> http://dl.dropbox.com/u/266880/Soma.pdf
>>>
>>> In the above paper, they mentioned
>>>
>>> the images of the neurons stained
>>>
>>>> for beta-III tubulin are first Gaussian filtered to suppress image
>>>> noise. They are background corrected using a top hat  
>>>> morphological filter
>>>> with a disk structuring element larger than the
>>>> size of the largest cell body. A morphological opening
>>>> with a structuring element of a diameter smaller than the cell
>>>> body but larger than all neurites permits suppressing neuritelike
>>>> structures.
>>>>
>>>>
>>> I am a little confused about the second and third step and wonder if
>>> anyone
>>> can help.
>>>
>>> Thank you.
>>>
>>> Qing
>>>
>>