Dark field particle counting on environmental mold slides — config advice before I build

Hi all — new here, first post.

I run a mold remediation company and I’m exploring whether the OpenFlexure can work as the starting imaging platform for a project I’m working on. The short version: I want to build a system that uses AI to count and classify mold spores on microscope slides from indoor air quality sampling cassettes, rather than sending them to a traditional lab.

The slides come from Air-O-Cell spore trap cassettes — standard in the indoor air quality industry. After collection, you open the cassette, mount the glass coverslip (roughly 10×15mm sample trace) on a standard 1×3" slide with lacto-phenol cotton blue stain, and examine it under a microscope. The particles of interest range from about 2–3 microns (Aspergillus/Penicillium conidia) up to around 30–40 microns for larger genera like Alternaria and Stachybotrys.

I want to use dark field rather than brightfield. The main reason is contrast — the small spores are lightly stained and can wash out under brightfield. Dark field also happens to be a different imaging approach than what existing commercial AI analysis services use, which matters for reasons I won’t bore you with. The ASTM D7391-20 standard for spore trap analysis explicitly mentions dark field and phase contrast as acceptable supplementary methods for looking at difficult particles, so it’s not outside the accepted methodology.

Right now I’m trying to figure out whether the OpenFlexure is the right starting point before I spend money on something more expensive. I have access to a 3D printer and I’m reasonably comfortable building things. I do not have a microscopy background — I’d be working with a certified spore analyst consultant to review image quality and label training data.

A few specific questions:

1. Dark field setup. I’ve seen reference in the forum to dark stop inserts for the condenser and the LED array approach in the Delta Stage paper. For this application — fixed slides, 40x, particles 2–40 microns — which approach would you actually recommend? Has anyone successfully done dark field on environmental particle slides specifically (as opposed to biological tissue or blood)?

2. Camera. I’m planning to use the 12MP Raspberry Pi HQ camera (IMX477) rather than the standard v2 module. I’ve seen at least one thread confirming this works. Is there anything non-obvious about fitting or configuring it — software settings, optics module changes, anything that bit you when you did it?

3. Objective for small spores. The concern I have is whether the standard 40x/0.65 NA plan objective can actually resolve a 2–3 micron Aspergillus spore well enough to distinguish it from debris or other small particles. Or does that size range really need 60x or 100x oil? I’d strongly prefer one objective that covers the full spore size range if that’s realistic.

4. Autofocus on sparse slides. Everything I’ve read about autofocus validation is on tissue sections or blood smears — high-contrast, dense samples. Air cassette slides are much sparser — mostly transparent particles on an adhesive background. Is autofocus reliable on that kind of low-contrast, low-density sample, or will it need manual help consistently?

5. Slide fit. The Air-O-Cell prep results in a thin glass coverslip mounted on top of a standard glass slide with mounting medium. The total thickness is a bit variable depending on how much medium was used. Any known issues with stage clearance or focus range on slides thicker than a standard single glass?

Thanks in advance — happy to share what I learn once I get into it. Seems like a different application than what most people here are working on but the underlying problem (automated particle counting from microscopy images) feels pretty aligned with the malaria and pathology work I’ve seen documented.

The LED array and condenser for the Delta stage could work with the normal microscope. The normal v7 microscope is far more stable and supported. We need to increase the number of modes and how they are described for the v7. This will probably happen for 7.1. v7 is taking a very long time to be fully released, but we are almost there!

Doing darkfield is possible, but far less turnkey and tested than brightfield.

You need a different optics module. And you will need an alpha release of the v3 software configured on a custom branch (note v3 software doesn’t yet support the delta stage.)

This is an image from a 40x microscope the smallest bars at the bottom are 2.19micron:

It depends what sort of detail you need on the 2–3 um spore.

Sparse autofocus is hard. There is some work in the pipline to improve this. Both thinking about being able to use the “focus figure of merit” reported from the picamera, and thinking about improving the autofocus motion algorithm.

Probably the easiest thing is the take a z-stack so you have a number of images separated by a small distance in z. These features are coming soonish to v3-alpha.

The issue you will have is darkfeild will work really badly with out current focussing metric as for speed we use jpeg compression size. This works ok in bright field, but very badly in darkfield. I would hope the autofocus figure of merit from the picamera would perform better, but this is untested.

The microscope is inverted, so we only look through the coverslip not the slide (as a standard 40x doesn’t have the working distance to work through a slide). It should be possible to get the z-axis low enough for uneven media. Depending how uneven you may want to mount the objective slightly low in the mounting slot.

I think my conclusion is that we are certainly a platform that could be used for developing such a technique, but we don’t have everything ready to go right now. So it depends on your lead time and expected development effort.

Out of curiosity, are you scanning the samples, and if so how big is the sample to be scanned?

Julian — thanks, this is really helpful and much more specific than I expected to get back so quickly.
To answer your question: yes, I’d be scanning the full sample trace. The Air-O-Cell deposit is roughly 10×15mm — an elliptical pattern on a standard 1×3" glass slide. So at 40x I’m looking at a fair number of fields to cover the full trace. I’d be interested to know whether the v7 stage travel range comfortably covers that area and what a realistic scan time looks like at that magnification.
On the development timeline — I’m not trying to go live quickly. I have some room for engineering overhead and I’m comfortable working with pre-release software if the alpha branch is reasonably stable. So the v3-alpha dependency isn’t a dealbreaker as long as I know what I’m getting into.
The autofocus point is the one I want to understand better before committing. If the current JPEG compression metric works badly in dark field, and the improved metric is coming in v3-alpha but untested — that’s a meaningful unknown for my use case, since I’d be running unattended overnight scans of multiple slides.
Which brings me to the question I’d most like your take on: given where things actually are right now, would you recommend starting in brightfield on the stable v7 release and switching to dark field once v7.1 and the improved autofocus metric are more settled — rather than trying to chase dark field on the current alpha? The dark field imaging mode matters to the project but it’s not so critical that it should block getting started. If brightfield on stable software gets me to generating useful training data in the near term, that seems like the better trade.
Is there a working dark field autofocus workaround already in use somewhere in the community, or is the z-stack approach in v3-alpha generzozuinely the best path right now?
Thanks again — this is already more useful than most of what I’d find in a spec sheet.
Tim

This is a really interesting application.

The nominal range of motion is 12×12mm, governed by the maximum safe flex angle of the plastic flexure hinges.

We recently found some issues with the main body design that mean that there is not a full 12×12 square available on the v7.0.0-beta5 release. This is discussed in the merge request !512. We are still testing it before the next release - the range of motion is now square, but is a little less than 12×12mm.

Updating configuration is relatively simple with the OpenFlexure Microscope. It is not designed to switch quickly during use, but it is designed to be able to print and replace parts for the different imaging possibilities without having to start from scratch. The LED array illuminator mentioned in the Delta Stage will also work on the standard Microscope - the relevant led_array_holder.stl is available if you ‘download every STL’ from the customisations page.

If you think that the OpenFlexure Microscope could be useful, then I would usually recommend starting with one of the standard brightfield versions because that is well tested and you can quite easily upgrade from there. As @julian says, for the software you would do well to look at the v3.0.0-alpha release as it will show you the improved scanning that we have now. To use the LED array for darkfield or pseudo-DIC you would need to use the v2 software.