How to average particle size distributions from images taken at vastly different scales? – Image Analysis

Hello – I’m currently analyzing supported Pt STEM images with Fiji. No-one in our group is any sort of expert, so the analysis is basically “draw shapes around relatively defined white regions that seem to be particles”, do this 1000 times, and present the distribution. Which is probably fine, but something that irks me and I don’t know a way around is the different scales of the images. One image is 800k mag, where as another is 12M mag we can see some 10 nm particles, while we see many single atoms or two or three atom clusters. Those atoms will be invisible at 800k mag, while if a 12M mag image has a 10nm particle in it, it’s only one cuz it takes up the whole picture.

I’m not sure what the best way to average is. People in the group just said that they don’t think about it, but surely someone has thought about this: Some sort of scale average, even if you have to do a little bootstrapping or something. A Google scholar and web of science search didn’t return much of anything.

Does anyone have any suggestions or “best-practice” ways of dealing with this? Or am I overthinking it?

Thanks!

I don’t think you really have the ability to average if your images are taken at different scales like that. At least not easily. If your high res images are such a small percentage of the area of a lower res image, anything that uniquely shows up in it needs to be modified for the relative volume. And yet, larger particles generally exclude smaller particles, so you would have to take into account the amount of free space, since you can’t just treat it as if an image 1000x smaller could be multiplied by 1000 times to account for the size change. After all, what if there isn’t that much free space?

In short, not sure there’s an easy way to average these things, you’d probably need to model it. Maybe someone else will have something better.

Thanks for the reply! I don’t do any of the imaging as our STEM doesn’t have the resolution. Are there ways to avoid this issue like choosing certain scales beforehand that I could request from the microscopist that you know of?