BrdU cell counting

[Josue]

Hello experts. I would like to know what method do you recommend for cell counting to measure cell proliferation in the SVZ when there are some clusters of uncountable cells?

[Olivier Burri]

Hello,

The simplest and usually still rather accurate way of doing this is to go for a threshold and measure the area covered by the cells.

Choosing the threshold: Try and use all the automatic methods imageJ has to offer and see if one of them matches a threshold that, in your expert opinion, captures the true BrdU area. Be careful using a manual threshold for each image, you might introduce bias. Maybe jumble your image names and offer a colleague a piece of cake for selecting the thresholds for you blindly.
 

You can then divide the area by the average size of a single BrdU positive cell. Measure a bunch of these individual cells to get a good statistic, get its standard deviation so that you can then get an estimate of the error you're making.

Eg:
Area: 5300 um2
Single cell mean: ~35um (measured over 100+ cells)
Single cell SD: 5.32um
Relative SD = 5.32/35 = 15.2%
Estimation on the number of cells: 5300/35 = 151.43 cells
Standard deviation : 151.43*0.152 = 23 cells

This usually gives you a good idea whether you can actually measure a difference between conditions. Once you have these data, you can use power-law calculations to see how many samples you'll need, for example.
http://biostat.mc.vanderbilt.edu/wiki/Main/PowerSampleSize


All the best,

Oli


Olivier Burri
Engineer - Image Processing
& Software Development
EPFL - SV - PTECH - PTBIOP


-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Josue
Sent: vendredi 10 janvier 2014 21:27
To: [hidden email]
Subject: BrdU cell counting

Hello experts.I would like to know what method do you recommend for cell counting to measure cell proliferation in the SVZ when there are some clusters of uncountable cells?



--
View this message in context: http://imagej.1557.x6.nabble.com/BrdU-cell-counting-tp5006086.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

[Josue]

Hello Olivier

Thank you very much for your kind response.
If you don't mind, I would like to ask you a couple of questions.

How many cells do you recommend measuring for a good estimation of number of cells? Would a 100 be ok?
If I estimate the area covered by cells, Will I need to set the scale to a given known value? I have a 1 mm rule in the magnification of the photos.
To measure the cells, Will I also need to have the scale set?

Thank you very much

[Olivier Burri]

Dear Josue,

> How many cells do you recommend measuring for a good estimation of
> number of cells? Would a 100 be ok?

As a rule of thumb, 100 cells could yield a good estimate. But this is very dependent on the kind of change you are expecting.
If between conditions your cells change in size but not in number, the approach I suggested is not ideal, as you'll think there are more cells when they actually just got bigger.
This is why a lot of people are satisfied with an area coverage for example, rather than actual cell numbers.

But if you have about 100 cells that are easily measurable per condition, then you can use that average to estimate the number inside the uncountable clusters. Just check that they don't change size between conditions.



> If I estimate the area covered by cells, Will I need to set the scale to a given
> known value? I have a 1 mm rule in the magnification of the photos.
> To measure the cells, Will I also need to have the scale set?

You can convert the pixel values to microns if you want to have the output directly scaled. Or you can do this all the way at the end.

Have all your measurements in pixels, when you estimate the total number of cells, it makes no difference as you're dividing an area in either pixels or microns with the average area of the cell you have either in pixels or microns. This is true as long as you do not have images at different magnifications.

To save you the hassle of putting everything to scale on each image, I'd suggest to do a cross-product at the end.

Otherwise:
If you have a 1mm rule, then you can draw a line profile on the rule and by using: "Analyze --> Set Scale...", you can easily calibrate your image.

As a final note, by the description of your images, there are a couple more questions I would have:

- What format are they? As they are RGB images, you'll need to use 'Color Deconvolution' to extract the DAB channel, but the results are less than optimal if you have a JPEG image. Make sure it's TIFF or some lossless format.

- You mention having a ruler on your image, so you should crop this out before you try to set a threshold, especially if you go for automated methods.

The easiest thing would be for you to send me or point me to one of your images so that we can check if the quality is good for starters.

All the best

Oli

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

[Gabriel Landini]

On Friday 17 Jan 2014 13:19:48 Burri Olivier wrote:
> > How many cells do you recommend measuring for a good estimation of
> > number of cells? Would a 100 be ok?
>
> As a rule of thumb, 100 cells could yield a good estimate. But this is very
> dependent on the kind of change you are expecting.

I am afraid that arbitrary numbers like "100" sound like "made up" numbers.

While 100 might be enough, really nobody knows what your data might look like
and you could do (statistically) better than that.

To find out how much data you need to collect, you need to apply a procedure
called "power analysis".
http://en.wikipedia.org/wiki/Statistical_power

That takes into account the variability of the data and the difference size
you are expecting to detect for a particular significance value. That way you
can obtain a statistically based estimate of the data size you need to
collect.
Without this you might be collecting too much data unnecessarily or too little
and assume there are no differences when actually they exist.

Cheers

Gabriel

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

[Josue]

Hello Olivier

Once more time, thank you very much for your kind help.

The format of the images are Tiff and please see the file attached in the link bellow of two images of the SVZ in two different conditions of the experiment. Please let me know if you need more information about the experiment

Thank you very much

https://www.dropbox.com/sh/ltyogpvrvww0k9g/pVjYOGZfSX

[Josue]

Thank you very much Gabriel.
The size of the cells is not supposed to change between conditions, therefore the number of the cells I need is just to estimate the final number of cells between conditions.

regards

Josue

[Gabriel Landini]

On Friday 17 Jan 2014 07:48:58 you wrote:
> The size of the cells is not supposed to change between conditions,
> therefore the number of the cells I need is just to estimate the final
> number of cells between conditions.

I do not mean the "size of cells", but the "size of the differences" you are
expecting to find in whatever you are measuring (number, sizes or any other
quantity).

Choosing arbitrary sample sizes is a bad idea when you have the opportunity of
doing an a priori power analysis.
Why? because at the end of experiments somebody *will* ask you to do a
posteriori power analysis and you risk facing that to find the differences you
are looking for needs way more data than you collected.

Like many things, ignoring the problem does not make it go away.

Cheers

Gabriel

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

[Josue]

Hi Gabriel!

Now I understand your recommendation!, I have used the PS2 to find out sample size and power calculation. Probably I will need to improve the quality of my images for using this method for cell counting.

Thank you very much for your response.

Josue