Obtaining number of particles vs Intensity data for evanescent wave images

imagej

#1

Hi,

I am working on detection of ceria nanoparticles by evanescent wave microscopy. To quantify the images obtained from this, I use both Imagej and DiaTrack softwares.

I am able to find the number of particles from Diatrack. However I am not able find it using imagej. I want to use imagej because it is more accurate than DiaTrack.

I have attached 2 images obtained from evanescent wave microscopy. I have also included the 3D surface plots obtained for these images from both DiaTrack and imagej. I am able to get intensity graph from DiaTrack but not from imagej. If you could let me know how to get the intensity plot from imagej it would be of great help. Thank you very much for your time and consideration.



#2

Would you like this procedure to be automated, to allow for fast analysis of many different images of the same type? Or is a manual workflow sufficient?

The images you shared have a white area around the edges, this causes problems with the surface plot etc. It can be removed, easily, at least manually, not sure how to automate it. It would be best if you could supply the raw images that dont have this border, is that possible?

Otherwise getting numbers of particles and the intensity of said particles is very doable and I can explain it to you! But the method depends on whether or not you want it to be automated or manual.

Hope I can help,
Sverre


#3

Respected Dr. Sverre,

Thank you very much for the reply.

The uploaded images of particles are actually snapshots of the original image. The size of original image is about 3 MB. I was not able to upload the original file as it took like 3 to 4 hours. I had no other go but to resort to this method of uploading a snapshot.

I would like this process to be automated, to allow fast analysis of many different images of the same type. It is also worth mentioning that these images have very few particles but there are some images with so many particles and that is where the actual challenge of counting is encountered. I have uploaded such images along with this message.

Also how can I find the size of the particles. Actually I am working on cleaning of these particles. For example, first time I will see like 3500 particles with a particular size distribution. After cleaning with the developed cleaning methodology I find that most of the brighter particles (either due to larger size or due to closeness to the interface) are cleaned. So this is another thing that is very important for my project.

Fortunately I was able to convert the image files to another format and the size reduced. So these are the original images and no white borders are seen.

Thank you very much for your time and consideration.



#4

For your second image and the images you have posted earlier, finding number of particles and measuring their intensity is a simple matter (search: Imagej particle analysis).

However, in the more crowded image you posted it looks like the particles are piled up, or forming aggregates, distinguishing these could be very hard, if possible at all. If they are indeed in a pile on top of one another you would need a 3d scan to separate objects?

But if you are sure that the particles are not stacked then maybe an automated threshold and watershed could suffice. Possibly a trained WEKA classifier for more robust analysis. Maybe you have some expert advice @etarena or @iarganda, can this densely populated image be realistically segmented?

In any case the analysis will not be perfect, what degree of accuracy do you require for these experiments? If you are just say, comparing the two images in your last post then we can of course plainly see the significant difference. Comparing two densely populated images to one another is a different matter…


#5

Respected Dr. Sverre,

Thank you very much for the reply.
I understood the particle analysis part and tried doing it for the images that I have uploaded but I am not able to get the result. It would be of great help if you can show how to obtain the number of particles vs intensity data for any of the images.

In the more crowded image, particles are actually piled up and you are right about that. I guess there is multilayer deposition of particles.

I understand that it is difficult to obtain accurate numbers for these images. It would be good if I can achieve close to 95% accuracy.

@etarena and @iarganda I would be indebted to you both if you could help me solve this problem of obtaining the desired data in imagej.

Thanks for your time and consideration.


#6

Hi @Akshay

Well… taking a look at your images, I agree with @Sverre. Your aggregates will be very difficult to segment. You would need a higher resolution image to more accurately segment these particles.

If that is not an option… you can look into using the Find Maxima tool in ImageJ to at least count those particles. Just be sure to apply the same exact settings to all the images you are counting - so test those settings on multiple conditions when you are setting things up - both high and low densities, etc.

Hope this helps a bit.

eta :slight_smile:


#7

Respected Arena,

Thank you very much for the reply.
I appreciate your help.
I totally agree with both you and Sverre.
As I said there is multilayer deposition and that is the reason we see such crowded images. Moreover my project involves studying the deposition of these particles on glass surface and removal of them using a developed cleaning methodology and that is the reason I am looking at quantification of these images so that after each clean step I can identify clearly how many particles have been removed.
I followed your suggestion to obtain the result from these images.
I have attached 2 images which shows how I processed the images using imagej.
I wanted to know if count = 383788 refers to the number of particles.
If we see the number of markers it looks so less but the magnitude of the count is too high. I really do not understand that.
Also for the same image using DiaTrack software I found that the number of particles is around 3000.

Would be grateful to you if you could clarify my doubts and let me know the exact procedure of analyzing these images.

Thanks for your time and consideration.


#8

Have you tried using the Trainable Weka Segmentation plugin? It would really help if you could upload an image at full resolution to somewhere like Dropbox or Google drive, together with an ideal result you want to obtain (segmentation or detection image).


#9

Respected Dr. Arganda,

@iarganda Thank you very much for the reply.

I have not tried that and will look into it.
Actually my main objective is to ‘obtain the number of particles vs intensity plot’ for each of these images.

I am not able to do this in imagej.

I have already uploaded the images at full resolution. Well, I am ready to upload more images through google drive or dropbox and share it with you (if you do not mind sharing your email id)

Thanks for your time and consideration.


#10

Just a quick reply to your question about this count number: This is just the number of pixels in your image, not the number of selected maxima.

In the ouput type > dropdown menu of find maxima, select count. A results table will appear with your maxima count.

This method works fine for counting, but for robust ROI selection and intensity measurements I would use the Trainable Weka Segmentation as Ignacio suggests. It is well documented, and you will find user guides on youtube :slight_smile:

I doubt anything can segment your multilayer images accurately though, you will probably have to get creative to solve this problem. After you reach a certain density of particles you will no longer be able to accurately measure an increase… How do other researchers work around this in similar experiments?

Sverre


#11

Respected Dr. Sverre,

Thank you very much for the reply.
I finally found out the method of obtaining count. I performed this analysis for an image and have uploaded the results
According to the table the maximum grayscale value is 235 but the maximum grayscale value through 3D plot is around 220.
Also, it looks like imagej is not helpful in this regard because through DiaTrack software I am able to directly obtain the number of particles vs intensity plot.
I will try to learn Trainable Weka Segmentation as suggested by Ignacio.

Well, I agree that it is difficult to segment multilayer images. I am not sure if anyone has worked on this before.

Thank you for your time and consideration.