If you want to count microglia in cell culture, automated image segmentation with Image J might be successful (particle analysis). However image segmentation in histological brain section in practice is much more challenging. First of all the reliability of your segmentation results relies on the signal to noise ratio of your immunohistological stainings. Another issue are cells which are directly overlying each other and are usually not distinguishable. In my experience currently existing image segmentation algorithms for brain section are insufficient and you should be quicker by manually counting your cells especially if you can apply a statistical sampling technique like stereology.
Microglial activation can be easily detected using an xCELLigence system (Roche). This system is based on impedance measurements and provides information on morphological differences. Real-time measurements also allow you to analyze when and for how long the microglia is activated.
If you want to count microglia in cell culture, automated image segmentation with Image J might be successful (particle analysis). However image segmentation in histological brain section in practice is much more challenging. First of all the reliability of your segmentation results relies on the signal to noise ratio of your immunohistological stainings. Another issue are cells which are directly overlying each other and are usually not distinguishable. In my experience currently existing image segmentation algorithms for brain section are insufficient and you should be quicker by manually counting your cells especially if you can apply a statistical sampling technique like stereology.
I completely agree with Peter. Stereology-based countings for brain sections are the best choice, since you spend time, reduce counting mistakes and (probably most important) this technique is widely accepted as the standard counting methods for most neuroscience journals and reviewers.
It depends on do you need absolute numbers or just a quantitative parameter that allows for comparison between different experimental groups?
In the first case your best option would be designed unbiased stereological counting (using fluorescence or EM ).
If you want to use relative parameter for comparisson, one way is to use some marker (e.g. IBA-1), segment the images and then quantify the positive area on every n-th slice. If you then calculate the IBA-1 positive area fraction , it will give you actually the positive volumetric fraction. Which can be compared between the test groups. If you know the tested volume you can also calculate the positive volume for each brain/brain fragment/tested nucleus etc. Of course you will not be able on 100% to prove if the differences are due to the changes in cell number or cell size. One way past this is to measure at the same time the average positive particle size. There are also very good morphological plugin collections (my personal favourites are Landini's morphological plugins ) that allow to get all kinds of particle form descriptive parameters, so this can help you even more. All of this can be automated using ImageJ macro recording or even ready plugins.
Keep in mind however that the most crucial step for an automated analysis is the segmentation , and for the segmentation the most important is the quality of image acquisition.
I recently did something similar but regarding astrocytes. For the identification I used GFAP. However I had access only to widefield epifluorescence microscope and thick slices so stereology was out of the question. Using ImageJ I could automate the entire process - all I had to do was to aquire the images and then gather and analyze the data.
I think (of course everybody is entitled to an opinion) that counting techniques and absolute numbers are bit "obsolete" when one counts immunolabeled structures. Simply said when it comes to immunolabeled structures there is no such thing as unbiased counting - as the Ig-Ab reaction depends on so many factors, and the different methods to "boost" the Ab signal (e.g. ABC) and there can also be unknown number of "hidden" in tissue "crevices" epitopes . So there always should be a control group and all data must be considered in comparison. That is why I think that designing and performing expensive, complex and time consuming analyzes just because it looks much better and sophisticated on paper, is a waste of time and money, if you can receive same results (answer whether there is difference or no difference between controls and test subjects) using simpler design.
If you face a difficulty in using the software itself, try to ask your question in ImageJ mailing list or search similar problem in http://imagej.1557.n6.nabble.com/.
Stoyan - I agree with your statement that no method is perfect (in particular when so vulnerable to the many steps of IHC)! However, microglia tend to alter shape and size through varying stages of activation (and across stage of development- so care must be taken with generalizations). As such, only doing area can be less sensitive - and when you count, you have the liberty of eye-balling the cells to make a further morphometric analysis. I acknowledge that, depending on the severity of the insult counting can be more sensitive (less severe), but with large focal lesions the confluence of the cells can cause difficulties in counts. As transmigration and microglial proliferation play a role in the magnitude of any response, both counts and total area can be complementary approaches to look at microglial activity. I have performed both with GFAP and find that the two do not always correlate - which in of itself is interesting.
@Stoyan Pavlov........We have serial sections of Albino mouse brain stained ith TTC (Mitochondrial stain). The damaged brain parts are unstained while the unaffected parts are stained red. We have scanned the stained brain slices now we want to measure the lesion sizes and uninfarcted parts of slices.
@Furhan Iqbal Basically the steps will be the following :
I would suggest that you open all your images belonging to one brain and add them to a stack (Image=>Stacks=>Images to stack) and save them .
Manual method: 1 Select on each image with the freehand selections tool the injured area Next you can run measure (With the Area, and area fraction options selected (in Analyze=>Set measurements ) Each measurement is added to a results table.
Semiautomatic method:
1. Segment - (threshold the images (Image=>Adjust=>Threshold) Be careful to threshold the unstained area as a "foreground" . You may have to perform some morphological operations (Closing) on your thresholded image until the infarction area is well separated and there is no other "noisy areas". If there are too many details and thresholding returns to many unstained areas that are hard to separate from the affected, you may try lowering the resolution of your images before the thresholding. 2. Run Measure with Area, and area fraction and limit to threshold options selected in Analyze=>Set measurements.
After you adjust all options, you can record your own macro with the help of (Plugins=>Macros=>Record). Tthere is a good ready to use macro that allows you to process entire folder.All you need is to paste your own macro and the pointed place and it is ready to use (http://rsb.info.nih.gov/ij/macros/BatchProcessFolders.txt) and automate this steps.
You may have to segment and measure the entire brain area against the background to measure the area of the entire brain slice. Once you have the Brain Slice Area and the Affected Area for each image in the series, you can easily calculate the mean area fraction, which is actually equal to the volumetric fraction. If you know the volume of the sectioned brains you can translate these numbers into affected volume for each brain.
As with any effort to take a qualitative endpoint to the next step to quantify, understanding the full nature of what you are measuring is critical. To take that further to design a method, or use a method, for automating this process takes even more understanding and experience with not only the cell type but also the methods and assumptions that underlie quantitation, e.g., cell counting. While a number of us could offer some level of advice, I would strongly encourage you to contact Dr. Peter Mouton, Professor of Stereology, in the Department of Pathology and Cell Biology at the University of South Florida School of Medicine. http://www.disector.com/Dr-Moutons-Biosketch-NIH_c_37.html. He has been working to develop a system that would allow for automated stereology. I think that you would find his work and his advice extremely helpful and authoritative. At a minimum he could offer you advice on the types of pitfalls that could arise and cautions needed for validating such an effort.
For many years I had the same feeling of Susan (i.e. NIH Image better than Image J), but when I discovered the FIJI project (http://fiji.sc/) I definitely left NIH Image for Image J. Now, there is a huge collection of plug in, macros etc... and the problem is how to navigate among them and found what is the most efficient for you.
The FIJI project (Fiji Is Just ImageJ) is a development platform. ImageJ2 will incorporate the best features of FIJI. Since NIH Image doesn't have nearly the extensibility as ImageJ / FIJI or ImageJ2 you would do best to adapt your counting technique with this program.
A few illustrative examples on how you can count cells using ImageJ and the ICTN plugin are listed here - http://www.youtube.com/watch?v=PqHFsmS1_JY and http://www.bioimage.ucsb.edu/automatic-nuclei-counter-plug-in-for-imagej . Of course you can use the built-in tools such as described above to a limited extend. A more full discussion and tutorial can be found here - http://www.unige.ch/medecine/bioimaging/tricks/imagejtutorials/CellCounting.pdf .
As you can see the best results are made when you have a specific stain for the microglia. This results in reducing the amount of information in the microscopy image, i.e. not all cells are labelled just the ones you are currently trying to count. If your staining and imaging technique have a large number of varying stained cells then the counting technique will have to incorparate a method of distinguishing the microglia from the other cells. In that case, you may need to filter the cells based on shapes, etc.
Two possibilities at analysis on shapes are shape descriptors and hull analysis. In this you would do well to look at Gary Chingas plugin here - http://www.gcsca.net/IJ/Shapes.html . A 3D hull analysis plugin has been written by Kris Sheets for some similar segmentation. The 3D shape plugin and documentation is available at https://sites.google.com/site/learnimagej/plugins/3d-shape .
Do not be discouraged at the large amount of options that you have in approaching your problem. A large number of people have been working steadily on many related problems and have found the ImageJ platform and it's variants a power place to build their efforts. I'm sure you will agree with them after some experience with the tools described above.
I have developed my own program to count microglia in IPL and OPL of the retina. I do not know if you are counting cells on the retina or in any other part so if you are interested just let me know and I can send you a aversion of the program (Matlab based).
Image J is super easy to use for quantification of either individual microglia, ramification across the tissues or just how darkly stained (OD) each microglia is. You can also use it to quantify individual size of microglia as an added bonus. Make sure first you only use the microglia stain and no counterstain, ensuring also there is minimum to no background.
Convert the picture to an 8 bit image, using the set threshold level highlight in red only the microglia manually, and repeat this operation on your photomicrographs so that you have the range of threshold that needs to be applied for measurements across your tissue set. We stain tissues in batches and ensure only crisply microliga are visible. Using the set measurements and objects tools in Image J you can set the program to measure the number of individual objects (in this scenario an object is a microliga that you define by size and circularity) and then also to measure the total staining (area) on the photomicrograph. By simple maths of subtracting the area cell bodies from the total staining (inclusive of ramified fibers) you can get some clever stats on the numbers of ramifications per microglia in tandem with simple measurements like optical density, which in a lab setting is really all you can get from ELISA or Western -based techniques in any event.
I highly recommend Image J, the software, especially as the program can be downloaded for free and used by any student from highschool to postgrad, allowing multiple users to verify counts on the same photomicrographs. The analysis is simple and sophisticated and can be applied to hundreds of photomicrographs, and adjusted based on the type of antibody (vascular, glial, neuronal or other) of interest. The main limitation of this technique is the quality and consitency of images that are used.. You just must ensure that the lighting microscope and mangification etc settings are equal when you take your original photographs, likewise fixing, staining and section thickness need to be the same.
We've used Image J for such purposes in pathological studies, although its not as old/established as stereology it is definitely more flexible and amenable to further development.
Just a word of caution with quantifying microglia with pixel based automated methods: how are you going to discern between the number of cells and their activiation state? Activated microglia are not as ramified as quiescent microglia, so it may be difficult to draw conclusions between recruitment of cells and their activation state. (Activated microglia will have fewer pixels since they're smaller) I encourage you to look into stereology - an unbiased method for quantification. If you are interested in determining the total population of microglia (or even the subpopuloations of activated and quiescent) the optical fractionator is a great method. You also mentioned, i think, a desire to measure the size of the infarct. That's very straightforward to do with stereology - using a probe called the area fraction fractionator. Quantification of cell number using single 2D image planes from tissue sections is biased - since you're counting parts of cells that are visible (rather than using rules to ensure that you have the opportunity to count a cell once and only once).
Just a note re how you remove bias in Image J, and avoid any double-counting, to address some issues suggested above.
Very simply stated to avoid the potential for double-counting you define objects (not pixels) that you wish to count has having a particular circularity and size before you do your cell counts. This mathematatical definitiion should define only the area of the larger individual neurons/glial cells. By instructing the software to only count the cells of this shape (identified further by using a threshold for the OD of staining) you can easily do cell counts which are unbiased.
Unfortunately not many people at the moment are aware of these additional functions in Image J and often the default is stereology. Likewise re activated or ramified microglia, you can use the software to calculate similar and also clusters of microglia, or even microglia specific to areas like perivascular regions in Image J.
To reiterate there is in all likelihood very little chance of counting the same cell twice if you have defined you are only counting cells which have as specific mean diameter (think of the area of a circle or other wise). And again, this software can be used by multiple users at the same time, which allows ready comparison for inter-rater reliability / sharing data. We have a few publications using Image J and also other not free-ware versions such as Image Pro Plus.
I have found that graduate students pickup the software very quickly, and also come up with clever ways of analyzing different changes on the photomicrographs.
I experience that developing, verifying , optimizing and running an image J quantification of a goo d staining take around 2h to 3h. Moreover i always count several picture manually to confirm the counting.
For picture with less that 50 events per pictures, blinding, opening and using cell counter plugin of image J take around 2 minutes per pictures.
So I won´t suggest image J plugin for less that 100 pictures, in that case I will prefer blinding and double counting.
See "Environmental impoverishment and aging alter object recognition, spatial learning, and dentate gyrus astrocytes" (DINIZ, D.G. et al 2010). Estereology method.
The DeadEasy Mito-Glia ImageJ Plugin is available from the Alicia Hidalgo Lab Page. Congratulations to this group for making there software available to others.
Look also in Prodanov D et al Automatic morphometry of synaptic boutons of cultured cells using granulometric analysis of digital images J Neurosci 2005
Having done thousands of manual reconstructions I decided there needed to be a better way! I have worked with some imaging engineers over the previous year to develop some software specifically to automate the process of microglial morphology analysis. Its very much in-house at this stage but if your interested please get in contact. I'd be happy to run a few images to look at feasibility ([email protected])
I think automation to count the cells might be tedious and sometimes ambiguous, given the differences in the shape, size and signal within the same section as well among sections.
I would prefer manual counting of cells in spite of any automated procedure.
However if the question is to assess the intensity, may be than it is feasible if signal to noise ratio is controlled.
There is no way to perform automatized unbiased counting. 3D stereology methods should be used for unbiased quantification purposes. Maybe optical dissector is the most commonly used.
You can either use an objective with a grid or use a Grid on ImageJ, there is a plug-in to do it I think. If not you can make your grid with a proper size. 50 µm could be OK.
If anyone happens to have interest in this thread anymore. I just published a method to automatically count microglial cells in the IPL and OPL of mice in Plos One. It works for healthy rats and in models of glaucoma too. Enjoy!!!