Flat-field correction of fluorescent images

[Rajendran, Megha]

Hi,

We use epifluorescence microscopy to study FRET using lanthanide metals.
We are just venturing into quantitative analysis and I am having some
issues correcting my images for uneven illumination.

(The image is of the metal solution in capillary for calibration and the
flat-field image was captured by putting a drop of lanthanide on a slide.)

I have a set of flat-field images at different gain. I was trying to
correct one image using imageJ as follows:
1. converted both my images to 32 bit (from 16/12 bit).
2. subtracted both images with average dark current noise image.
2. normalized my flatfield image by dividing with the max signal.
3. divided my original image with the normalized flatfield image.
When I check the plot profile across the diagonal, the field is even. Then
i measure the mean, min and max gray value. I compared this with the
values before correction.
                   Area           Mean            Min     Max
Before flat-field: 1310720 1337.228 699.752 1900.651
After flat-field:  1310720 1899.424 661.128 8839.022

I don't know why the values are so high after correction. I was wondering
if I am missing a step here.

(I also tried the correction using the original image, by blurring it and
then using it as flat-field image. I still see the high values. sometimes
the max values are as high as 60000.)

Let me know if you need more information
Thanks,
Megha

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

[Nico Stuurman]

Hi Megha,

On Nov 22, 2012, at 10:27 AM, Rajendran, Megha wrote:

> I have a set of flat-field images at different gain. I was trying to correct one image using imageJ as follows:
> 1. converted both my images to 32 bit (from 16/12 bit).
> 2. subtracted both images with average dark current noise image.
> 2. normalized my flatfield image by dividing with the max signal.

I think that you will want to normalize your flatfield image with the average (not max) signal.  That should also  lower the values after flatfielding.

> 3. divided my original image with the normalized flatfield image.
> When I check the plot profile across the diagonal, the field is even. Then i measure the mean, min and max gray value. I compared this with the
> values before correction.
>                   Area           Mean            Min     Max
> Before flat-field: 1310720 1337.228 699.752 1900.651
> After flat-field:  1310720 1899.424 661.128 8839.022
>
> I don't know why the values are so high after correction. I was wondering
> if I am missing a step here.

Best,

Nico

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

[Rajendran, Megha]

Thanks Nico and John,

Sorry for the delayed reply, I wanted to try all the suggestions before replying.

Using the mean intensity solved the problem with high values after flatfield
correction. I was still getting very high min and max values, so I tried smoothing
the Flat-field image using guassian blur (radius 20). These together got me some
sensible values. I am getting negative minimum intensity after subtracting dark
current noise, I am not sure if it matter but wanted to know if I can avoid it (I
switch off all the lights in the room to take these images and average by stacking
and average Z-project).

My values are as follows: (mean intensity; minimum; maximum intensity)
raw image: 1336.853; 501; 4092
substract dark current noise: 1044.771, -1835.300, 2681.300
Divide flat-field (normalized by dividing mean intensity): 1045.247; -1789.488;
3014.886
Without guassian blur, the mean intensity remains same but the min and max intensity
goes crazy.

I have another problem, I plot-profile the 2 diagonals before and after flat-field
correction. one diagonal shows the uneven illumination pattern and corrects
significantly after correction. The other diagonal seems fine to start with but gets
uneven after correcting. I guess it is before my reference images are not right for
my sample. I have tried getting reference images twice at different conditions
(using slide/coverslip and in rectangular capillary, my samples are fluorophore
solution in capillaries), but both the flat-field images have same pattern and it is
different from my samples light pattern. I was wondering if I can do
pseudo-correction using the image itself
(http://www.uhnres.utoronto.ca/facilities/wcif/imagej/image_intensity_proce.htm -
section 7.1.2); has anyone done this before?

Hope I am not confusing things up. Thanks for all the help

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

[pang]

Dear Megha,

I don't think that your dark current noise is 1044.771. It should be slight greater than your minimum intensity (501). I had published a paper and wonder if it can give you some information you need.

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2818.2011.03581.x/abstract



Zhengyu

Zhengyu Pang, Ph.D.
Biochemistry and Bioanalytics Laboratory
Diagnostic and Biomedical Technologies
GE Global Research, K1-5B37A
One Research Circle
Niskayuna, NY 12309
T: 518-387-4015
F: 518-387-7765




Thanks Nico and John,

Sorry for the delayed reply, I wanted to try all the suggestions before replying.

Using the mean intensity solved the problem with high values after flatfield correction. I was still getting very high min and max values, so I tried smoothing the Flat-field image using guassian blur (radius 20). These together got me some sensible values. I am getting negative minimum intensity after subtracting dark current noise, I am not sure if it matter but wanted to know if I can avoid it (I switch off all the lights in the room to take these images and average by stacking and average Z-project).

My values are as follows: (mean intensity; minimum; maximum intensity) raw image: 1336.853; 501; 4092 substract dark current noise: 1044.771, -1835.300, 2681.300 Divide flat-field (normalized by dividing mean intensity): 1045.247; -1789.488;
3014.886
Without guassian blur, the mean intensity remains same but the min and max intensity goes crazy.

I have another problem, I plot-profile the 2 diagonals before and after flat-field correction. one diagonal shows the uneven illumination pattern and corrects significantly after correction. The other diagonal seems fine to start with but gets uneven after correcting. I guess it is before my reference images are not right for my sample. I have tried getting reference images twice at different conditions (using slide/coverslip and in rectangular capillary, my samples are fluorophore solution in capillaries), but both the flat-field images have same pattern and it is different from my samples light pattern. I was wondering if I can do pseudo-correction using the image itself (http://www.uhnres.utoronto.ca/facilities/wcif/imagej/image_intensity_proce.htm - section 7.1.2); has anyone done this before?

Hope I am not confusing things up. Thanks for all the help

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

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

[carlo bianco]

hello,
I'd like to know if is possible to set a 100 point grid
for point counting, I've got the grid plugin, but it sets

the grid with areas, not with number of points.
Thank you in advance for your help
bye
carlo bianco
bologna university
Italy



________________________________
 Da: "Pang, Zhengyu (GE Global Research)" <[hidden email]>
A: [hidden email]
Inviato: Lunedì 26 Novembre 2012 21:35
Oggetto: Re: Flat-field correction of fluorescent images
 
Dear Megha,

I don't think that your dark current noise is 1044.771. It should be slight greater than your minimum intensity (501). I had published a paper and wonder if it can give you some information you need.

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2818.2011.03581.x/abstract 



Zhengyu

Zhengyu Pang, Ph.D.
Biochemistry and Bioanalytics Laboratory
Diagnostic and Biomedical Technologies
GE Global Research, K1-5B37A
One Research Circle
Niskayuna, NY 12309
T: 518-387-4015
F: 518-387-7765




Thanks Nico and John,

Sorry for the delayed reply, I wanted to try all the suggestions before replying.

Using the mean intensity solved the problem with high values after flatfield correction. I was still getting very high min and max values, so I tried smoothing the Flat-field image using guassian blur (radius 20). These together got me some sensible values. I am getting negative minimum intensity after subtracting dark current noise, I am not sure if it matter but wanted to know if I can avoid it (I switch off all the lights in the room to take these images and average by stacking and average Z-project).

My values are as follows: (mean intensity; minimum; maximum intensity) raw image: 1336.853; 501; 4092 substract dark current noise: 1044.771, -1835.300, 2681.300 Divide flat-field (normalized by dividing mean intensity): 1045.247; -1789.488;
3014.886
Without guassian blur, the mean intensity remains same but the min and max intensity goes crazy.

I have another problem, I plot-profile the 2 diagonals before and after flat-field correction. one diagonal shows the uneven illumination pattern and corrects significantly after correction. The other diagonal seems fine to start with but gets uneven after correcting. I guess it is before my reference images are not right for my sample. I have tried getting reference images twice at different conditions (using slide/coverslip and in rectangular capillary, my samples are fluorophore solution in capillaries), but both the flat-field images have same pattern and it is different from my samples light pattern. I was wondering if I can do pseudo-correction using the image itself (http://www.uhnres.utoronto.ca/facilities/wcif/imagej/image_intensity_proce.htm - section 7.1.2); has anyone done this before?

Hope I am not confusing things up. Thanks for all the help

--
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

[Rajendran, Megha]

Hi Zhengyu,

Thanks for the paper.

My dark current noise values are : Mean: 292.082; Min: 191.500; Max: 4092

That value is of original images after dark current noise subtraction. Do these values make more sense?

Thanks
Megha

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

[Gabriel Landini]

On Tuesday 27 Nov 2012 00:08:00 you wrote:
> My dark current noise values are : Mean: 292.082; Min: 191.500; Max: 4092

That means that when there is no light hitting the sensor you get a reading of
4092 for some pixel(s)?
Can you explain step by step, how do you actually capture the dark current
image? Thanks

Gabriel

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

[Rajendran, Megha]

Hi Gabriel,

I shut off the led light source, I switch off all the lights in the room and then capture 10 images for each gain and exposure setting. I thought the 4092 is from hot-pixels.  

Megha

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

[Jacqueline Ross]

Hi Carlo,

You could try modifying the DrawRandomDots.txt macro if you want random placement of the points (http://rsbweb.nih.gov/ij/macros/DrawRandomDots.txt).

It puts 25 random dots on an image but you could change that to 100 and modify the size of the dots so that they aren't too large depending on your image resolution..

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 1142, NEW ZEALAND

Tel: [phone]
Fax: [phone]

http://www.fmhs.auckland.ac.nz/sms/biru/

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of carlo bianco
Sent: Tuesday, 27 November 2012 9:57 a.m.
To: [hidden email]
Subject: 100 point grid for point counting

hello,
I'd like to know if is possible to set a 100 point grid for point counting, I've got the grid plugin, but it sets

the grid with areas, not with number of points.
Thank you in advance for your help
bye
carlo bianco
bologna university
Italy



________________________________
 Da: "Pang, Zhengyu (GE Global Research)" <[hidden email]>
A: [hidden email]
Inviato: Lunedì 26 Novembre 2012 21:35
Oggetto: Re: Flat-field correction of fluorescent images
 
Dear Megha,

I don't think that your dark current noise is 1044.771. It should be slight greater than your minimum intensity (501). I had published a paper and wonder if it can give you some information you need.

http://onlinelibrary.wiley.com/doi/10.1111/j.1365-2818.2011.03581.x/abstract 



Zhengyu

Zhengyu Pang, Ph.D.
Biochemistry and Bioanalytics Laboratory Diagnostic and Biomedical Technologies GE Global Research, K1-5B37A One Research Circle Niskayuna, NY 12309
T: 518-387-4015
F: 518-387-7765




Thanks Nico and John,

Sorry for the delayed reply, I wanted to try all the suggestions before replying.

Using the mean intensity solved the problem with high values after flatfield correction. I was still getting very high min and max values, so I tried smoothing the Flat-field image using guassian blur (radius 20). These together got me some sensible values. I am getting negative minimum intensity after subtracting dark current noise, I am not sure if it matter but wanted to know if I can avoid it (I switch off all the lights in the room to take these images and average by stacking and average Z-project).

My values are as follows: (mean intensity; minimum; maximum intensity) raw image: 1336.853; 501; 4092 substract dark current noise: 1044.771, -1835.300, 2681.300 Divide flat-field (normalized by dividing mean intensity): 1045.247; -1789.488;
3014.886
Without guassian blur, the mean intensity remains same but the min and max intensity goes crazy.

I have another problem, I plot-profile the 2 diagonals before and after flat-field correction. one diagonal shows the uneven illumination pattern and corrects significantly after correction. The other diagonal seems fine to start with but gets uneven after correcting. I guess it is before my reference images are not right for my sample. I have tried getting reference images twice at different conditions (using slide/coverslip and in rectangular capillary, my samples are fluorophore solution in capillaries), but both the flat-field images have same pattern and it is different from my samples light pattern. I was wondering if I can do pseudo-correction using the image itself (http://www.uhnres.utoronto.ca/facilities/wcif/imagej/image_intensity_proce.htm - section 7.1.2); has anyone done this before?

Hope I am not confusing things up. Thanks for all the help

--
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

[Gabriel Landini]

> I shut off the led light source, I switch off all the lights in the room and
> then capture 10 images for each gain and exposure setting. I thought the
> 4092 is from hot-pixels.  

Most microscopes have a manual shutter that prevents any light reaching the
camera. That avoids switching on and off the light.

Did you take a look at the distribution (histogram) of hot pixels?
If you get such a high number for dark current I wonder if the gain or range
in your camera is set too high. Or maybe the camera is set to auto-gain (some
have it), in which case should be switched off. What camera is this?
Dark current is usually very small unless you integrate for long times.

Cheers

Gabriel

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

[Rajendran, Megha]

Hi Gabriel,

I dark current noise image was for high gain and exposure (g8 clk20). I am calibrating the fluorescence intensity for all the settings. I did a histogram and there are like 13 pixels of 4092 value. I tried it for lower gain value and there is still pixels at 4092 just fewer ( g7 clk10 - 5, g7 clk3 - 1). Should I remove the hot pixels by denoise before using the image?

The camera is MEGA-10 from stanford photonics.


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

[Gabriel Landini]

> I dark current noise image was for high gain and exposure (g8 clk20). I am
> calibrating the fluorescence intensity for all the settings. I did a
> histogram and there are like 13 pixels of 4092 value. I tried it for lower
> gain value and there is still pixels at 4092 just fewer ( g7 clk10 - 5, g7
> clk3 - 1). Should I remove the hot pixels by denoise before using the
> image?
 
> The camera is MEGA-10 from stanford photonics.

Shouldn't the minimum of the darkfield be very near 0 instead of 191? I guess
that you have too much gain.
Another source of problems is that the camera gets hot as you go along the day
and the bias increases. So the darkfield should be captured before you are
intending to snap the images.

Yes, you will get erroneous image values for these 13 bright pixels as they
output the maximum value, not a biased one (maybe they do, but it eventually
saturates due to long exposure or they might get binned at 4096 by the 8 bit
histogram?).

If you know exactly where they are, you can ignore them, or replace their
value with the average of their neighbours. Many consumer cameras do this,
with a median filter. In the wiki there is a macro to shows how to do this
only for the hot pixels.

I think that for darkfield images (unlike brightfield), you do not compute the
transmittance (i.e. dividing by the background and rescaling), but just
subtract the backgound image. You should be able to capture without a specimen
or estimated by several methods (curve fitting to background points or
morphological filtering).  Please correct me if I am wrong.

Cheers
Gabriel

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

[Rajendran, Megha]

Thanks Gabriel,

I am going to redo my reference images, so I will try taking the dark current noise the first thing this time.
Regarding the last para of darkfield images, I think you are right about subtracting the background and not dividing it. I plan to correct all the images for uneven-illumination and then subtract them with a background image.

By the way do you have any idea about the pseudo-correction for flat-field, as mentioned in the link I have in one of the posts?

Megha

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