Segmentation troubleshooting

[Allison Hall]

Hi,

 

I am at present using Image J to quantify functional fluorescently

> stained blood vessels present in mouse tissues from treated animals.

> We have previously utilized the OTSU plug-in followed by the despeckle


> filter for the analysis of these mouse tissues, but found that we were


> unable to utilize this plug-in for the analysis of mouse tumours. I

> believed that is was due to the lack of "information" within the

> tumour section (due to its disorganized structure, and the lack to

> fluorescence due to vasculature shutdown). The trial that I am

> analyzing at present is now seeing the tumour section issue, where due


> to vasculature shutdown OTSU is dramatically under thresholding the

> images. Please find attached an example of a control original

> composite image and an image which has almost complete vasculature

> shutdown, and the respective OTSU outputs. Note: these are both kidney


> sections

>  

> We have had a play around with other filters and found that on a small


> subset of affected and nonaffected sections and found that Dynamic

> Threshold 1b appears to threshold to mirror the information seen on

> the original composite image. What is the mask size?? Note: I used

> mask size of 1 in generating the images attached.

 

Control kidney.bmp

https://www7.sendthisfile.com/d.jsp?t=ZuSm75qT6H6vnVUWsNDeF7wF

Mean dynamic control kidney.bmp

https://www5.sendthisfile.com/d.jsp?t=X6TT76sbTpfc0z2yEDIkIIUX

Median dynamic control kidney.bmp

https://www7.sendthisfile.com/d.jsp?t=YP2jPKUH1tDwRNDRWPPIzQvX

OTSU control kidney no threshold.bmp

https://www2.sendthisfile.com/d.jsp?t=T9mrfT6EQbj1rtXADCociIl0

OTSU control kidney with threshold.bmp

https://www6.sendthisfile.com/d.jsp?t=slNIa1XOU13OQBMwAfWI3iCO

treated kidney.bmp

https://www3.sendthisfile.com/d.jsp?t=MpLepGjfSWtt6CIAihVR8MQz

Mean dynamic treated kidney.bmp

https://www7.sendthisfile.com/d.jsp?t=0opyINN6ajmITjGWqutDSM4d

Median dynamic treated kidney.bmp

https://www7.sendthisfile.com/d.jsp?t=JX0PebhsC9meeD48Ix5ttFvg

OTSU treated kidney no threshold.bmp

https://www3.sendthisfile.com/d.jsp?t=OlwBibxYmKaPTyYYuksLg9sp

OTSU treated kidney with threshold.bmp

https://www4.sendthisfile.com/d.jsp?t=lUbn47BQ7pWSYckHmfTD8CqH

 

Note: These images are only available for 6 days after being uploaded

 

Also is it more accurate to OTSU threshold with a selection around the
section of interest or just the entire composite image, from playing
around the threshold is set higher when the selection is applied prior
to thresholding than not.

>  

> If you could please have a look at the images and hopefully you will

> have some suggestions, like is there a more suitable filter or plugin

> we could use, etc

>  

 

 

Allison Hall

Research Assistant

Cancer Drug Discovery Group

 

Bionomics Limited

31 Dalgleish Street

Thebarton SA 5031

Phone: [phone]

Fax: [phone]

Email: [hidden email] <mailto:[hidden email]>

Website: www.bionomics.com.au <http://www.bionomics.com.au>

ABN 53 075 582 740

 

[Karsten Rodenacker]

Hi,
you show interesting examples.
Allow me some remarks:
1. Seemingly you have used images which were originally stored in  
jpeg format with relatively low quality. I think you should improve  
the quality considerably if the vessel structure is of interest.
2. Vessel structure of the control seems similar to the vessel  
structure in the treated case INSIDE some lets call it hot (cold?)  
regions or patches.
3. For characterizing structures never use mosaics from beginning!

For differentiation of tumour and non tumor regions it should be  
clear which clues should be used or are of importance e.g.: micro-
structure(texture), macro structure and/or isolated events ...:

I recommend math morphology:  (see plugin graymorphology)
microstructure: grayscale tophat transformation radius 2, circular):  
original minus open(radius 2); threshold
macrostructure: vessel free or vessel rare zones: inversion of  
closure (radius 4-5, circular) of the microstructure; threshold
isolated events: difficult in control, bright spots in treated.  
Seemingly in control there are spots like in treated, slightly  
larger, but with finer vessel structure.

Good luck,
KR

Am 19.01.2007 um 01:10 schrieb Allison Hall:

> Image J to quantify functional fluorescently

Karsten Rodenacker
-------------------------------------------------------------------- :-)
GSF - Forschungszentrum     Institute of Biomathematics and Biometry
D-85758 Oberschleissheim    Postfach 11 29
Karsten.Rodenacker_AT_gsf.de | http://ibb.gsf.de/
http://ibb.gsf.de/homepage/karsten.rodenacker/
Tel: [phone] | FAX: ..193401

[Allison Hall]

Thanks for the response,

We are actually interested in quantifying the amount of perfusion
(viable blood vessels) within the tumour or tissue section, not
specifically looking at the individual structures. Is your suggestion
applicable to this?

Our images were originally saved in bmp format, and the composite image
is also saved as a bmp. You mention never to use mosaics, is this also
the case for quantifying vessels??

Allison Hall
Research Assistant
Cancer Drug Discovery Group
 
Bionomics Limited
31 Dalgleish Street
Thebarton SA 5031
Phone: [phone]
Fax: [phone]
Email: [hidden email]
Website: www.bionomics.com.au
ABN 53 075 582 740

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
Karsten Rodenacker
Sent: Friday, 19 January 2007 7:23 PM
To: [hidden email]
Subject: Re: Segmentation troubleshooting

Hi,
you show interesting examples.
Allow me some remarks:
1. Seemingly you have used images which were originally stored in  
jpeg format with relatively low quality. I think you should improve  
the quality considerably if the vessel structure is of interest.
2. Vessel structure of the control seems similar to the vessel  
structure in the treated case INSIDE some lets call it hot (cold?)  
regions or patches.
3. For characterizing structures never use mosaics from beginning!

For differentiation of tumour and non tumor regions it should be  
clear which clues should be used or are of importance e.g.: micro-
structure(texture), macro structure and/or isolated events ...:

I recommend math morphology:  (see plugin graymorphology)
microstructure: grayscale tophat transformation radius 2, circular):  
original minus open(radius 2); threshold
macrostructure: vessel free or vessel rare zones: inversion of  
closure (radius 4-5, circular) of the microstructure; threshold
isolated events: difficult in control, bright spots in treated.  
Seemingly in control there are spots like in treated, slightly  
larger, but with finer vessel structure.

Good luck,
KR

Am 19.01.2007 um 01:10 schrieb Allison Hall:

> Image J to quantify functional fluorescently

Karsten Rodenacker
-------------------------------------------------------------------- :-)
GSF - Forschungszentrum     Institute of Biomathematics and Biometry
D-85758 Oberschleissheim    Postfach 11 29
Karsten.Rodenacker_AT_gsf.de | http://ibb.gsf.de/
http://ibb.gsf.de/homepage/karsten.rodenacker/
Tel: [phone] | FAX: ..193401

[Karsten Rodenacker]

Am 22.01.2007 um 00:53 schrieb Allison Hall:

>
> We are actually interested in quantifying the amount of perfusion
> (viable blood vessels) within the tumour or tissue section, not
> specifically looking at the individual structures. Is your suggestion
> applicable to this?
>
Tophat by a given radius detects after appropriate threshold all  
bright structures with diameter smaller then 2*radius+1. If these are  
the vessels you are interested in, the area/measurementarea is a good  
estimate for perfusion, or better this is an estimate of the volume  
density of vessels.  Try to find literature under "stereology,  
perfusion, kidney etc.". For measurement the areas (vessel, field) of  
different images have to be summed up, hence no mosaicing is necessary.

Tophat transformation can be considered as a sieve or as a background  
correction, where the opened image (in a certain way smoothed image)  
is considered as background, allowing to apply one threshold for  
detection of the (tophat size) objects.

> Our images were originally saved in bmp format, and the composite  
> image
> is also saved as a bmp. You mention never to use mosaics, is this also
> the case for quantifying vessels??
>
In that case I ask me where the tessel structure comes from (small  
tessels)? I don't know bmp format, I know these artefacts only from  
jpeg images. My recommendation not to use mosaics results from the  
difficulties to standardize the different neighbouring images in a  
way that quantification (thresholding!) is possible. In your examples  
you have gray value differences and overlay (stitching) artefacts.  
Mosaics are e.g. sensefull if you try to quantify spatial  
relationships or to get an overview.

KR

> -----Original Message-----
> From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of
> Karsten Rodenacker
> Sent: Friday, 19 January 2007 7:23 PM
> To: [hidden email]
> Subject: Re: Segmentation troubleshooting
>
> Hi,
> you show interesting examples.
> Allow me some remarks:
> 1. Seemingly you have used images which were originally stored in
> jpeg format with relatively low quality. I think you should improve
> the quality considerably if the vessel structure is of interest.
> 2. Vessel structure of the control seems similar to the vessel
> structure in the treated case INSIDE some lets call it hot (cold?)
> regions or patches.
> 3. For characterizing structures never use mosaics from beginning!
>
> For differentiation of tumour and non tumor regions it should be
> clear which clues should be used or are of importance e.g.: micro-
> structure(texture), macro structure and/or isolated events ...:
>
> I recommend math morphology:  (see plugin graymorphology)
> microstructure: grayscale tophat transformation radius 2, circular):
> original minus open(radius 2); threshold
> macrostructure: vessel free or vessel rare zones: inversion of
> closure (radius 4-5, circular) of the microstructure; threshold
> isolated events: difficult in control, bright spots in treated.
> Seemingly in control there are spots like in treated, slightly
> larger, but with finer vessel structure.
>
> Good luck,
> KR
>
> Am 19.01.2007 um 01:10 schrieb Allison Hall:
>
>> Image J to quantify functional fluorescently
>
> Karsten Rodenacker
> -------------------------------------------------------------------- :
> -)
> GSF - Forschungszentrum     Institute of Biomathematics and Biometry
> D-85758 Oberschleissheim    Postfach 11 29
> Karsten.Rodenacker_AT_gsf.de | http://ibb.gsf.de/
> http://ibb.gsf.de/homepage/karsten.rodenacker/
> Tel: [phone] | FAX: ..193401

Karsten Rodenacker
-------------------------------------------------------------------- :-)
GSF - Forschungszentrum     Institute of Biomathematics and Biometry
D-85758 Oberschleissheim    Postfach 11 29
Karsten.Rodenacker_AT_gsf.de | http://ibb.gsf.de/
http://ibb.gsf.de/homepage/karsten.rodenacker/
Tel: [phone] | FAX: ..193401