Non-uniform background removal for dot blot integrated pixel density

[bmb242]

Hi everyone. I apologize for the long message but I am overwhelmed with many projects and this is causing a lot of issues!

I am completely new to image analysis and have been given the task of analyzing some protein dot blots that I did for my research. I know very little about this so I have a couple of questions. I have scanned a piece of paper that contains 7 sample membranes, each membrane containing a different protein concentration (and for each membrane, there are 4 spots of same protein concentration so I can get an average). I need to produce a standard curve of the average integrated pixel density of each membrane 4-spot set against known protein concentration.

My issue is this. Each sample has non-uniform background noise which I need to correct for and I am not sure how to do this. I have seen a couple of methods but they are ambiguous and a bit confusing and am not sure if I can apply them. From what I understand, I roughly need to do this: measure integrated pixel density of the four spots per membrane, find the average (easy). Subtract background from that to get a proper reading. My issue lies with the background subtraction since it is non uniform and I do not know the best method.

So my questions are essentially this:

1. I am now thinking I need to separate this scanned image into 7 images, each containing ONLY the membrane being analyzed and nothing else (paper it is attached to, other membranes etc). I believe this is necessary for proper background subtraction. Right now, the image is a piece of white paper boxed off in 7 sections with labels and the membrane sample in each. Is this correct? I guess I could crop each membrane into it's own file, but am afraid that will alter image quality, although I know cropping in this way is not manipulation.

2. For subtraction...I used ROI manager to select all four spots for Sample 1, for example, and found mean integrated pixel density. To subtract the background should I....take a few random ROI from the background, average them, and use that mean density as background? Do the size spots need to be equal to my sample spot sizes? If so, should I leave out any background areas where it was clear something occurred such as poor washing (lets say one corner is very dark, but does not interfere with sample spots) or is this manipulation?

If it would help I can upload one sample image for clarity or produce a model. Just let me know.

I am just generally confused with how to do this. I have spent hours and hours looking online trying to figure it out, but I have no definitive answers and the LAST thing I want to do is manipulate my data in an inappropriate way.

I really appreciate the help. I am in bioinformatics and image analysis is not my area and I am having a lot of trouble. I need to complete this so I can finish reports and move on.

Thank you so much.

[bmb242]

Still really struggling with this unfortunately. Hoping to get some insight.


~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~

Hi everyone. I apologize for the long message but I am overwhelmed with many projects and this is causing a lot of issues!

I am completely new to image analysis and have been given the task of analyzing some protein dot blots that I did for my research. I know very little about this so I have a couple of questions. I have scanned a piece of paper that contains 7 sample membranes, each membrane containing a different protein concentration (and for each membrane, there are 4 spots of same protein concentration so I can get an average). I need to produce a standard curve of the average integrated pixel density of each membrane 4-spot set against known protein concentration.

My issue is this. Each sample has non-uniform background noise which I need to correct for and I am not sure how to do this. I have seen a couple of methods but they are ambiguous and a bit confusing and am not sure if I can apply them. From what I understand, I roughly need to do this: measure integrated pixel density of the four spots per membrane, find the average (easy). Subtract background from that to get a proper reading. My issue lies with the background subtraction since it is non uniform and I do not know the best method.

So my questions are essentially this:

1. I am now thinking I need to separate this scanned image into 7 images, each containing ONLY the membrane being analyzed and nothing else (paper it is attached to, other membranes etc). I believe this is necessary for proper background subtraction. Right now, the image is a piece of white paper boxed off in 7 sections with labels and the membrane sample in each. Is this correct? I guess I could crop each membrane into it's own file, but am afraid that will alter image quality, although I know cropping in this way is not manipulation.

2. For subtraction...I used ROI manager to select all four spots for Sample 1, for example, and found mean integrated pixel density. To subtract the background should I....take a few random ROI from the background, average them, and use that mean density as background? Do the size spots need to be equal to my sample spot sizes? If so, should I leave out any background areas where it was clear something occurred such as poor washing (lets say one corner is very dark, but does not interfere with sample spots) or is this manipulation?

If it would help I can upload one sample image for clarity or produce a model. Just let me know.

I am just generally confused with how to do this. I have spent hours and hours looking online trying to figure it out, but I have no definitive answers and the LAST thing I want to do is manipulate my data in an inappropriate way.

I really appreciate the help. I am in bioinformatics and image analysis is not my area and I am having a lot of trouble. I need to complete this so I can finish reports and move on.

Thank you so much.

~~~~~~~~~~~~~~~~~~~~~~~~~~~
~~~~~~~~~~~~~~~~~~~~~~~~~~~

[GDC]

Is the nonuniformity from the illumination or from the camera?  If you
have access to the camera the images came from you can collect a dark
field image and subtract this from your images to remove non uniformity
from the sensor.
  Greg

On 4/3/2015 3:26 PM, bmb242 wrote:

> Still really struggling with this unfortunately. Hoping to get some insight.
>
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~
> ~~~~~~~~~~~~~~~~~~~~~~~~~~
>
> Hi everyone. I apologize for the long message but I am overwhelmed with many
> projects and this is causing a lot of issues!
>
> I am completely new to image analysis and have been given the task of
> analyzing some protein dot blots that I did for my research. I know very
> little about this so I have a couple of questions. I have scanned a piece of
> paper that contains 7 sample membranes, each membrane containing a different
> protein concentration (and for each membrane, there are 4 spots of same
> protein concentration so I can get an average). I need to produce a standard
> curve of the average integrated pixel density of each membrane 4-spot set
> against known protein concentration.
>
> My issue is this. Each sample has non-uniform background noise which I need
> to correct for and I am not sure how to do this. I have seen a couple of
> methods but they are ambiguous and a bit confusing and am not sure if I can
> apply them. From what I understand, I roughly need to do this: measure
> integrated pixel density of the four spots per membrane, find the average
> (easy). Subtract background from that to get a proper reading. My issue lies
> with the background subtraction since it is non uniform and I do not know
> the best method.
>
> So my questions are essentially this:
>
> 1. I am now thinking I need to separate this scanned image into 7 images,
> each containing ONLY the membrane being analyzed and nothing else (paper it
> is attached to, other membranes etc). I believe this is necessary for proper
> background subtraction. Right now, the image is a piece of white paper boxed
> off in 7 sections with labels and the membrane sample in each. Is this
> correct? I guess I could crop each membrane into it's own file, but am
> afraid that will alter image quality, although I know cropping in this way
> is not manipulation.
>
> 2. For subtraction...I used ROI manager to select all four spots for Sample
> 1, for example, and found mean integrated pixel density. To subtract the
> background should I....take a few random ROI from the background, average
> them, and use that mean density as background? Do the size spots need to be
> equal to my sample spot sizes? If so, should I leave out any background
> areas where it was clear something occurred such as poor washing (lets say
> one corner is very dark, but does not interfere with sample spots) or is
> this manipulation?
>
> If it would help I can upload one sample image for clarity or produce a
> model. Just let me know.
>
> I am just generally confused with how to do this. I have spent hours and
> hours looking online trying to figure it out, but I have no definitive
> answers and the LAST thing I want to do is manipulate my data in an
> inappropriate way.
>
> I really appreciate the help. I am in bioinformatics and image analysis is
> not my area and I am having a lot of trouble. I need to complete this so I
> can finish reports and move on.
>
> Thank you so much.
>
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
> ~~~~~~~~~~~~~~~~~~~~~~~~~~~
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012338.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[bmb242]

Greg,

I have attached a crude example. The four circles are the samples I need to quantify. The rectangles are the membranes used. The white around the rectangles is the piece of paper I had the sample membranes attached to when I scanned them. Everything around the circles is background, and as I simply showed, it is non-uniform. Each background is unique and non uniform to each membrane. This is what I need to correct for. Find background for each membrane sample...subtract it so I can find average of pixel density of four spots on that membrane...then move to next one. My first question refers to cropping the whole image into multiple images, each with one membrane ONLY, so it does not show that white piece of paper or the other membranes.

The nonuniformity is essentially improper washing of the membranes during the assay that could not be fixed. Essentially, I am using a specific detection method which makes the sample spots appear due to a precipitation reaction. This also produces background around the samples where it should theoretically not (but this is the real world so background happens as you know). The membranes are originally white. Ideally, upon completion of the assay, the membranes would remain white except for the four sample dots, which would be varying shades of grey based on the protein concentration. Unfortunately, there is background, and it is not uniform. In other words, some background areas are darker than others on the same membrane, potentially from improper washing steps, or indirect binding to the membrane. Because of this I can not perform a simple background subtraction since pixel density varies across the background. The step where the samples are scanned has no effect on the background, it is the sample membranes themselves that have background from the actual assay steps.

Hope that answers your question. I really appreciate the reply!

Here is the crude example:

http://i.imgur.com/AWSzXVH.png

<quote author="GDC">
Is the nonuniformity from the illumination or from the camera?  If you
have access to the camera the images came from you can collect a dark
field image and subtract this from your images to remove non uniformity
from the sensor.
  Greg

[GDC]

If I understand you correctly, you are trying to quantify spots on
section of a non uniform scanned image?  If I am correct in my
understanding, I think you want to normalize each section to itself.  I
am assuming since these are scanned images that they are relatively low
resolution (I am also assuming that you don't have the ability to get
digital photos of the membranes).  If you crop your scans as you suggest
you can then export the resulting image as a text image and manipulate
it with your math package of choice, you could probably even get away
with a spreadsheet package such as calc(A text image is a numeric array
of the grayscale values that form the image).  If you choose this route
make sure you use text image and not text file.  Once you get each image
squared away within itself you can normalize image to image in order to
compare one to another.

There is an ebook on using Image J called "Image Processing with Image
J" by Perez and Pascau that is a very good entry level text

I hope this is helpful

On 4/3/2015 10:57 PM, bmb242 wrote:

> Greg,
>
> I have attached a crude example. The four circles are the samples I need to
> quantify. The rectangles are the membranes used. The white around the
> rectangles is the piece of paper I had the sample membranes attached to when
> I scanned them. Everything around the circles is background, and as I simply
> showed, it is non-uniform. Each background is unique and non uniform to each
> membrane. This is what I need to correct for. Find background for each
> membrane sample...subtract it so I can find average of pixel density of four
> spots on that membrane...then move to next one. My first question refers to
> cropping the whole image into multiple images, each with one membrane ONLY,
> so it does not show that white piece of paper or the other membranes.
>
> The nonuniformity is essentially improper washing of the membranes during
> the assay that could not be fixed. Essentially, I am using a specific
> detection method which makes the sample spots appear due to a precipitation
> reaction. This also produces background around the samples where it should
> theoretically not (but this is the real world so background happens as you
> know). The membranes are originally white. Ideally, upon completion of the
> assay, the membranes would remain white except for the four sample dots,
> which would be varying shades of grey based on the protein concentration.
> Unfortunately, there is background, and it is not uniform. In other words,
> some background areas are darker than others on the same membrane,
> potentially from improper washing steps, or indirect binding to the
> membrane. Because of this I can not perform a simple background subtraction
> since pixel density varies across the background. The step where the samples
> are scanned has no effect on the background, it is the sample membranes
> themselves that have background from the actual assay steps.
>
> Hope that answers your question. I really appreciate the reply!
>
> Here is the crude example:
>
> http://i.imgur.com/AWSzXVH.png
>
>
> Is the nonuniformity from the illumination or from the camera?  If you
> have access to the camera the images came from you can collect a dark
> field image and subtract this from your images to remove non uniformity
> from the sensor.
>    Greg
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012342.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[bmb242]

Yes, if you look at the crude model I uploaded you can see the four membranes, each with four spots on them. For example...sample 1 has four spots that were applied using a 1mg/ml protein solution. I need to average the pixel densities of those spots to get one number. That number then correlated to the 1mg/ml concentration and can be a point on my standard curve. Then I move to the next membrane of 0.1mg/ml, 0.01mg/ml, 0.001mg/ml etc. Each background per membrane is non-uniform and different than the other membranes. So I guess I need to correct background to normalize to itself, then normalize to each other? This is way over my head right now so I guess I'll look into that book.

I have various forms of the image but the highest resolution is 174.2 MB, 26736x16608 and is a .tiff.

You don't think I can simply use ROI manager to get pixel density of the spots, then select random parts of the background to average to get the background since it is non-uniform? I am unsure of what math package you refer to.

I am sorry, I really am not trying to ask anyone to do this for me, I am just super confused with all of this. I will continue to do reading.

Thanks!

GDC wrote

If I understand you correctly, you are trying to quantify spots on
section of a non uniform scanned image?  If I am correct in my
understanding, I think you want to normalize each section to itself.  I
am assuming since these are scanned images that they are relatively low
resolution (I am also assuming that you don't have the ability to get
digital photos of the membranes).  If you crop your scans as you suggest
you can then export the resulting image as a text image and manipulate
it with your math package of choice, you could probably even get away
with a spreadsheet package such as calc(A text image is a numeric array
of the grayscale values that form the image).  If you choose this route
make sure you use text image and not text file.  Once you get each image
squared away within itself you can normalize image to image in order to
compare one to another.

There is an ebook on using Image J called "Image Processing with Image
J" by Perez and Pascau that is a very good entry level text

I hope this is helpful

On 4/3/2015 10:57 PM, bmb242 wrote:
> Greg,
>
> I have attached a crude example. The four circles are the samples I need to
> quantify. The rectangles are the membranes used. The white around the
> rectangles is the piece of paper I had the sample membranes attached to when
> I scanned them. Everything around the circles is background, and as I simply
> showed, it is non-uniform. Each background is unique and non uniform to each
> membrane. This is what I need to correct for. Find background for each
> membrane sample...subtract it so I can find average of pixel density of four
> spots on that membrane...then move to next one. My first question refers to
> cropping the whole image into multiple images, each with one membrane ONLY,
> so it does not show that white piece of paper or the other membranes.
>
> The nonuniformity is essentially improper washing of the membranes during
> the assay that could not be fixed. Essentially, I am using a specific
> detection method which makes the sample spots appear due to a precipitation
> reaction. This also produces background around the samples where it should
> theoretically not (but this is the real world so background happens as you
> know). The membranes are originally white. Ideally, upon completion of the
> assay, the membranes would remain white except for the four sample dots,
> which would be varying shades of grey based on the protein concentration.
> Unfortunately, there is background, and it is not uniform. In other words,
> some background areas are darker than others on the same membrane,
> potentially from improper washing steps, or indirect binding to the
> membrane. Because of this I can not perform a simple background subtraction
> since pixel density varies across the background. The step where the samples
> are scanned has no effect on the background, it is the sample membranes
> themselves that have background from the actual assay steps.
>
> Hope that answers your question. I really appreciate the reply!
>
> Here is the crude example:
>
> http://i.imgur.com/AWSzXVH.png
>
>
> Is the nonuniformity from the illumination or from the camera?  If you
> have access to the camera the images came from you can collect a dark
> field image and subtract this from your images to remove non uniformity
> from the sensor.
>    Greg
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012342.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[bmb242]

Okay so I made an attempt using a plug-in for non-uniform background correction: http://www.cs.unc.edu/~cquammen/imagej/nonuniform_background_removal.html

I cropped the whole image into separate tiff files of each membrane. I used the same sized ROI circle the entire time for both samples and background samples.

Method for each membrane:

1. Select four randomly dispersed background circles
2. Run plug-in to correct for background
3. Select all four spots in produced corrected image (I used a smaller circle than the size of the actual spot)
4. Find average integrated density of those four spots combined

Unfortunately, the data does not go as I expect it do. From what I understand, darker spots should have smaller integrated density values. Mine kind of seem all over the place...not sure if I need to correct against each other?? Either way I think the data should make more sense.

Does this indicate a problem with my data, or with my method? It should theoretically be linear along with the known concentrations.

Thank you!!!!

[GDC]

"You don't think I can simply use ROI manager to get pixel density of the
spots, then select random parts of the background to average..."

This is one way to do it and if the backgrounds were uniform to each
other, pretty much exactly what you need to do. To normalize between
images you would select the exact same spots in all of the images. You
do this by defining multiple ROI's for the background.  The ROI's will
define an XY location to pull the data from. You want the ROI locations
to be the same for each image.  In order for this to work, the cropped
images must be exactly the same size and referenced to the point within
the image.

The problem comes from the the fact that, as I understand it, the
backgrounds are not only non uniform within the image, but also across
images.  The method I suggested will allow you to normalize pixel by
pixel.  Any math software package should be able to this manipulation (
MatLab, Mathcad, SMath, Mathmatica, Octave, SciLab, etc.).  If each
cropped image were smaller than say 640x480 you might be able to use any
common spreadsheet software such as Calc or Excel.

How close is your signal gray level to the background gray levels? If
they are very close, withing ~10 gray levels or so, then you will need a
sophisticated method to sort out the data.  If the background and signal
are clearly differentiated then the method you describe should work very
well.

I hope this is helpful


On 4/6/2015 2:49 PM, bmb242 wrote:

> Yes, if you look at the crude model I uploaded you can see the four
> membranes, each with four spots on them. For example...sample 1 has four
> spots that were applied using a 1mg/ml protein solution. I need to average
> the pixel densities of those spots to get one number. That number then
> correlated to the 1mg/ml concentration and can be a point on my standard
> curve. Then I move to the next membrane of 0.1mg/ml, 0.01mg/ml, 0.001mg/ml
> etc. Each background per membrane is non-uniform and different than the
> other membranes. So I guess I need to correct background to normalize to
> itself, then normalize to each other? This is way over my head right now so
> I guess I'll look into that book.
>
> I have various forms of the image but the highest resolution is 174.2 MB,
> 26736x16608 and is a .tiff.
>
> You don't think I can simply use ROI manager to get pixel density of the
> spots, then select random parts of the background to average to get the
> background since it is non-uniform? I am unsure of what math package you
> refer to.
>
> I am sorry, I really am not trying to ask anyone to do this for me, I am
> just super confused with all of this. I will continue to do reading.
>
> Thanks!
>
>
> GDC wrote
>> If I understand you correctly, you are trying to quantify spots on
>> section of a non uniform scanned image?  If I am correct in my
>> understanding, I think you want to normalize each section to itself.  I
>> am assuming since these are scanned images that they are relatively low
>> resolution (I am also assuming that you don't have the ability to get
>> digital photos of the membranes).  If you crop your scans as you suggest
>> you can then export the resulting image as a text image and manipulate
>> it with your math package of choice, you could probably even get away
>> with a spreadsheet package such as calc(A text image is a numeric array
>> of the grayscale values that form the image).  If you choose this route
>> make sure you use text image and not text file.  Once you get each image
>> squared away within itself you can normalize image to image in order to
>> compare one to another.
>>
>> There is an ebook on using Image J called "Image Processing with Image
>> J" by Perez and Pascau that is a very good entry level text
>>
>> I hope this is helpful
>>
>> On 4/3/2015 10:57 PM, bmb242 wrote:
>>> Greg,
>>>
>>> I have attached a crude example. The four circles are the samples I need
>>> to
>>> quantify. The rectangles are the membranes used. The white around the
>>> rectangles is the piece of paper I had the sample membranes attached to
>>> when
>>> I scanned them. Everything around the circles is background, and as I
>>> simply
>>> showed, it is non-uniform. Each background is unique and non uniform to
>>> each
>>> membrane. This is what I need to correct for. Find background for each
>>> membrane sample...subtract it so I can find average of pixel density of
>>> four
>>> spots on that membrane...then move to next one. My first question refers
>>> to
>>> cropping the whole image into multiple images, each with one membrane
>>> ONLY,
>>> so it does not show that white piece of paper or the other membranes.
>>>
>>> The nonuniformity is essentially improper washing of the membranes during
>>> the assay that could not be fixed. Essentially, I am using a specific
>>> detection method which makes the sample spots appear due to a
>>> precipitation
>>> reaction. This also produces background around the samples where it
>>> should
>>> theoretically not (but this is the real world so background happens as
>>> you
>>> know). The membranes are originally white. Ideally, upon completion of
>>> the
>>> assay, the membranes would remain white except for the four sample dots,
>>> which would be varying shades of grey based on the protein concentration.
>>> Unfortunately, there is background, and it is not uniform. In other
>>> words,
>>> some background areas are darker than others on the same membrane,
>>> potentially from improper washing steps, or indirect binding to the
>>> membrane. Because of this I can not perform a simple background
>>> subtraction
>>> since pixel density varies across the background. The step where the
>>> samples
>>> are scanned has no effect on the background, it is the sample membranes
>>> themselves that have background from the actual assay steps.
>>>
>>> Hope that answers your question. I really appreciate the reply!
>>>
>>> Here is the crude example:
>>>
>>> http://i.imgur.com/AWSzXVH.png
>>>
>>>
>>> Is the nonuniformity from the illumination or from the camera?  If you
>>> have access to the camera the images came from you can collect a dark
>>> field image and subtract this from your images to remove non uniformity
>>> from the sensor.
>>>     Greg
>>>
>>>
>>>
>>> --
>>> View this message in context:
>>> http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012342.html
>>> Sent from the ImageJ mailing list archive at Nabble.com.
>>>
>>> --
>>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>> --
>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012372.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[GDC]

I have not found integrated density to be a useful measurement in my
work.  Integrated density is the mean gray level of the ROI multiplied
by the area of the ROI.  Your problem may lie in the background
differences between the images.  If you just look at the mean graylevel
for your dark spots does it make more sense?

On 4/6/2015 4:44 PM, bmb242 wrote:

> Okay so I made an attempt using a plug-in for non-uniform background
> correction:
> http://www.cs.unc.edu/~cquammen/imagej/nonuniform_background_removal.html
>
> I cropped the whole image into separate tiff files of each membrane. I used
> the same sized ROI circle the entire time for both samples and background
> samples.
>
> Method for each membrane:
>
> 1. Select four randomly dispersed background circles
> 2. Run plug-in to correct for background
> 3. Select all four spots in produced corrected image (I used a smaller
> circle than the size of the actual spot)
> 4. Find average integrated density of those four spots combined
>
> Unfortunately, the data does not go as I expect it do. From what I
> understand, darker spots should have smaller integrated density values. Mine
> kind of seem all over the place...not sure if I need to correct against each
> other?? Either way I think the data should make more sense.
>
> Does this indicate a problem with my data, or with my method? It should
> theoretically be linear along with the known concentrations.
>
> Thank you!!!!
>
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012376.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[Gabriel Landini]

On Monday 06 Apr 2015 13:44:09 bmb242 wrote:
> Okay so I made an attempt using a plug-in for non-uniform background
> correction:
> http://www.cs.unc.edu/~cquammen/imagej/nonuniform_background_removal.html

I haven seen what your images look like but the above plugin will fit a plane
or a cubic polynomial to the background and use that fitted, so I guess that
you would need more than 4 samples to fit a cubic polynomial surface to the
background.

If the light source is uneven, then you might be able to compensate it to some
extent by using images of the original light source (maybe now it is too
late?) and then compute the transmittance, same as in brightfield microscopy:
http://imagejdocu.tudor.lu/doku.php?id=howto:working:how_to_correct_background_illumination_in_brightfield_microscopy

Note that the a-posteriori methods mentioned in the link above (which include
the method using surface fitting like the plugin you tried) make assumptions
about your images which might not be correct.

> It should theoretically be linear along with the known concentrations.

Not sure what you mean here, but if you are talking about the relation between
concentration and optical density, then I think you need to calibrate your
image (after the background correction) with an optical density ladder,
captured at the same time of the scan.

Hope this is useful.
Cheers

Gabriel

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

[Gabriel Landini]

On Tuesday 07 Apr 2015 00:06:42 Gabriel Landini wrote:
> I haven seen what your images look like but the above plugin will fit a
> plane or a cubic polynomial to the background and use that fitted, so I
[...]
I meant "use that fitted surface to correct the image,..."

Cheers

Gabriel

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

[bmb242]

Thanks a ton for both of your insights.

There is no light source in this procedure. The background and non uniformity are on the actual membranes, it is not a reflection, or optical issue. The actual physical membrane, when looked at in person with the naked eye, has background since the reaction solutions used also react with the membrane (although less than the samples, hence the spots being darker than the background).

Yes, the backgrounds are all different and all non uniform.

I will run another trial and generate an entirely new set of samples. This time I will make sure all spots are in the EXACT same spot on the membrane between all of the membranes, and the cropped images the exact same size so I can try and do the method you mentioned previously going pixel by pixel using the same XY locations on each membrane as samples and background.

Gabriel, the images look just like the lame MS paint example I posted above. I can crop the whole image down to get separate images of the membrane only, which I have done and used in the method above. It is a flat nitrocellulose membrane.

It seems mean grey value is also all over the place.

Thank you for your patience and help everyone

[Gabriel Landini]

On Tuesday 07 Apr 2015 10:27:44 you wrote:
> There is no light source in this procedure.

Hi,
I would be really surprised if that was the case. How is the image created? If
you scanned it, there is a light source somewhere.

> The actual physical membrane, when looked at in person with the naked
> eye, has background since the reaction solutions used also react with the
> membrane (although less than the samples, hence the spots being darker than
> the background).

In that case you have a bigger problem than 'background illumination'. If your
lighting is not uneven (or the source of background variability) it might be
difficult or just not possible to make a distinction of what is 'background'
and what is the spot. Just imagine that there might be a 'dark background
patch' on top of the spot. How do you decide the contribution of each to the
detected optical density?

The background correction methods we discussed before do not resolve this
issue, only the uneven illumination. They are based on the assumption that if
you know how much light there was originally (without sample, i.e. a
brightfield image) then you can measure how much less light there is now in
the actual image. If you say that the background is in the membrane itself,
then this is additional to the background illumination problem.

Can I suggest that you reduce one of your images into a manageable size and
post it somewhere for people to see. Sorry to say that the MS pain examples
are not good to understand what is going on. Maybe somebody will have an idea
to help you.

Cheers
Gabriel

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

[bmb242]

Okay here is an example. These are two of the membranes. You can see the non-uniform "background" (caused by the reagents I use in my research during washing and visualizing steps) and can see how that background is different between the two membranes.

http://i.imgur.com/feXTJy2.jpg?1

[GDC]

Thanks for the actual image  it is very helpful!  Are you scanning the
membranes as grayscale images? Have you scanned them as color images?  
If so you might try splitting the rgb channels to see if one channel has
more usable info.
Some of your problem may be from the light source in the scanner. If you
have the ability to take photos in a loss less format such as .RAW or
uncompressed tiff with a continuous light source and split the channels
or filter the image you may be able to eliminate much of your non
uniformity. You can split the image into individual rgb channels under
Image/Color.

Good Luck

On 4/9/2015 12:00 PM, bmb242 wrote:

> Okay here is an example. These are two of the membranes. You can see the
> non-uniform "background" (caused by the reagents I use in my research during
> washing and visualizing steps) and can see how that background is different
> between the two membranes.
>
> http://i.imgur.com/feXTJy2.jpg?1
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012425.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[bmb242]

They are scanned in black/white I believe.

I'm just having a hard time understand how the scanner light will some how eliminate the black spots caused by the reactions. They are on the actual membranes and don't have to do with the light. Am I really confused here?

GDC wrote

Thanks for the actual image  it is very helpful!  Are you scanning the
membranes as grayscale images? Have you scanned them as color images?  
If so you might try splitting the rgb channels to see if one channel has
more usable info.
Some of your problem may be from the light source in the scanner. If you
have the ability to take photos in a loss less format such as .RAW or
uncompressed tiff with a continuous light source and split the channels
or filter the image you may be able to eliminate much of your non
uniformity. You can split the image into individual rgb channels under
Image/Color.

Good Luck

On 4/9/2015 12:00 PM, bmb242 wrote:
> Okay here is an example. These are two of the membranes. You can see the
> non-uniform "background" (caused by the reagents I use in my research during
> washing and visualizing steps) and can see how that background is different
> between the two membranes.
>
> http://i.imgur.com/feXTJy2.jpg?1
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012425.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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

[GDC]

The colors in scanned images  and most photos are comprised of a
combination of red, green and blue pixels.  Your stains may be more or
less sensitive one of these channels.  When you scan in grayscale, you
average these channels to form the image.

If you can collect images with an actual camera you will have much more
flexibility in terms of illumination and or collection wavelength
through the use of filters.  You may be able to find a wavelength that
accentuates your signal.

On 4/12/2015 9:34 PM, bmb242 wrote:

> They are scanned in black/white I believe.
>
> I'm just having a hard time understand how the scanner light will some how
> eliminate the black spots caused by the reactions. They are on the actual
> membranes and don't have to do with the light. Am I really confused here?
>
>
> GDC wrote
>> Thanks for the actual image  it is very helpful!  Are you scanning the
>> membranes as grayscale images? Have you scanned them as color images?
>> If so you might try splitting the rgb channels to see if one channel has
>> more usable info.
>> Some of your problem may be from the light source in the scanner. If you
>> have the ability to take photos in a loss less format such as .RAW or
>> uncompressed tiff with a continuous light source and split the channels
>> or filter the image you may be able to eliminate much of your non
>> uniformity. You can split the image into individual rgb channels under
>> Image/Color.
>>
>> Good Luck
>>
>> On 4/9/2015 12:00 PM, bmb242 wrote:
>>> Okay here is an example. These are two of the membranes. You can see the
>>> non-uniform "background" (caused by the reagents I use in my research
>>> during
>>> washing and visualizing steps) and can see how that background is
>>> different
>>> between the two membranes.
>>>
>>> http://i.imgur.com/feXTJy2.jpg?1
>>>
>>>
>>>
>>> --
>>> View this message in context:
>>> http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012425.html
>>> Sent from the ImageJ mailing list archive at Nabble.com.
>>>
>>> --
>>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>> --
>> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
>
>
>
> --
> View this message in context: http://imagej.1557.x6.nabble.com/Non-uniform-background-removal-for-dot-blot-integrated-pixel-density-tp5012260p5012446.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

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