Hi Paris,

As you suggest, laser diffraction reports an equivalent spherical diameter. To a good approximation, this diameter relates to the sphere of equivalent volume. In addition, the distributions reported are weighted according to the volume of sample within each size class.

In comparing imaging with diffraction, you need to account for the volume weighting. By default, imaging analysis returns a number-based distribution whereas laser diffraction is volume-based. Converting from one to the other can introduce errors, especially if you do not sample a large amount of material using your imaging system. Typically, number based methods report a smaller particle size than laser diffraction, as they see the 1000s of small particles within a sample but are statistically unlikely to see a low number of larger particles. In laser diffraction, a small number of large particles contain a significant volume, so are much more likely to be detected. Laser diffraction will, however, be insensitive to changes in number of very fine particles present within the size distribution.

There is another consideration in comparing laser diffraction and imaging, and that relates to particle orientation. Plate-like particles (such as clays) will flow align when measured in laser diffraction rather than tumbling randomly through the measurement system. The technique will therefore tend to report a size based on the largest cross-section of the particle, rather than a diameter which is based on the overall particle volume. This can complicate the process of comparing methods.

In conclusion: consider sampling first. If you can sample a large number of particles using imaging, you may be able to obtain good comparisons with laser diffraction. Once this is considered, you’ll be able to more confidently explore which diameter from imaging gives the best correlation with your laser diffraction data.

Hope this is useful information.

All the best,

Paul

Hello Paul,

thanks for the information. I guess there is not golden rule of how much is enough. Depending on the specimen, it can vary a lot. From 50 particles up to a couple of hundreds.

But from what I understood, even in the best case scenario, comparisons should be made with caution.

Many thanks

Paris

Using Image J measuring tool one can calibrate the size of particles. But, it is should be done after calibrating the scale used in our image w.r.t to pixel size in Image J. One should randomly select the particles repeatedly for number of times to achieve better results.

Hello Anandha,

yes I did calibrate the image in ImageJ before using it. 

I guess you suggest using a random subsample for many times? I think it can easily be done in R.

Thanks for the reply

Paris

Hi Paris,

You are correct - it is sample dependant so there is no golden rule. It is also dependant on which parmaeter you are trying to compare also (something close to the median size or something close to the edges of the distribution).

I do not like direct promotion of Malvern stuff in answers to forums like this. However, we have a recorded webinar which considers comparisons between laser diffraction and imaging. The link should be attached to this comment below. Hope it is useful. I also hope it is generic enough to apply to any imaging system.

All the best,

Paul

http://www.malvern.com/en/support/events-and-training/webinars/W091119LaserDiffractionMasterclass7.aspx

Hello Paris,

I'm not 100% sure what particles are you working with...can you express ther shape and size by some parameters? Circularity for example - ratio of the perimeter of a circle with the same area as the particle divided by the perimeter of the actual particle image.

From my experience, particles with circularity over 0.78 can be treated as spherical and LDM comprison with microscopic imaging is done quite well.

If you can measure particle area (pixel) and express it as CE diameter in microscopic imaging it should be comparable with SE in LDM as long as it has over 0.78 circularity.

If they are more irregular then I can suggest using some wet cell to make particles float when imaging them with microscope. Then the orientation should be more random - hence more comparable to LDM.

Czarek

Article Influence of the Sand Particle Shape on Particle Size Distri...

Article Particle size and shape characteristics of kemerburgaz quart...

Thank you Paul and Cezary for the additional information.

I will have a careful look and decide what to do with my samples.

Hello Paris,

can you upload an example image? There might be an opportunity for automated image analysis with our algorithm

http://www.sciencedirect.com/science/article/pii/S0098135412001603

Hello Sebastian,

I attach a picture of my one of specimens. As you can the grains are agglomerated, hence I don't know if your algorithm will be able to recognise the different particles. Some of the images from in the paper look similar, so you think it could work?

Thanks for your time

First my thought was, that what you could do is to reconstruct it in 3d with fib/sem - but it's whole new story there I imagine ;)

Did you think about doing it other way around - try to make such agglomerates from particles in sediments or destroy those agglomerates to analyse separate particles? How hard is it to break it?

Then you could say something about differences between agglomerates (surface roughness?) or particles.

To make separate particles from that photo in Matlab or Labview you need some patterns of separate particles or some parameter which make them separate. And you don't have any pattern to match now, because you want to compare it with your possible pattern - sediment particles. Latter is rather doubtful. The thing is that no matter what you do with this 2d picture it will be approximation of approximation because you don't see most of the particle...

Anyway, you can try to tweak with Sebastian code or try Labview Vision Development module (it has trial period) - Vision Assistant to be more precise-  to try to match patterns quite easily - hopefully it would be easier to test it briefly than tweak with Matlab or ImageJ - doesn't mean it will work at all though. Maybe it will find some sediment particles in this agglomerate if you provide many search patterns...

I will try what will work - so, there are some very, very big and some very tiny particles - Are you interested in both tails of the distribution?

Hello Cezary,

I would prefer not to destroy the specimens for now, thus I will try and play around with Vision Assistant and see if it gets me anywhere. 

Hello Sebastian, the ideal would be to count as many particles as possible from all sizes. Because the very small ones (e.g. >5 μm or >10μm) might be too difficult to count, I guess the larger ones would be more important.

Hello Paris,

we developed an automatized algorithm in order to determine the primary particle size of nanoparticles from TEM pictures.

You can find the imageJ plugin and the reference of the publication at this address : 

http://www.coria.fr/spip.php?article910

Hello Alexandre,

interesting work. Thanks for the info. I will play with plugin you provided, and I ll see where it gets me.