Microscopes meet Artificial Intelligence

Microscopes and the Abbe Limit

Microscopes come in all shapes and sizes. Some work with visible light, others with electrons, some even work with atomic force — effectively feeling their way around and object. It all depends on what you want to look at. Some microscopes are great for looking at live animals, others are designed for looking at very small things like proteins and, well, viruses.


Thankfully though, some clever folk came up with Single Molecule Localisation Microscopy or SMLM, and specifically, Stochastic Optical Reconstruction Microscopy(or STORM for short). This form of microscopy is part of a family of techniques known as Super Resolution Microscopy. So called because they can cheat their way past the Abbe limit by making a few sacrifices.

Enter Deep learning

So, we’ve successfully got some 2D images of a particular object, like a protein complex (in our case, a centriole). How do we turn these 2D images into some sort of 3D model? Well, this is all quite easy you might think. We’ve had 3D reconstruction for ages, like photogrammetry or Shape from shade or even existing AI based approaches. Sadly, this doesn’t work too well with our problem for a few reasons.

Experiments and statistics in pandemic times

I’ve been working away at this project for almost 3 years. For the first year I was remote. As soon as I got back to the UK, the pandemic hit, so being back in the office was not an option. I basically had to figure out a way to share my results remotely. Fairly easy you’d think right?

Publishing your results

It’s important to share your work — possibly the most important thing! Peer review is a big part of the science, though the current publishing model has some serious problems. Before I get to these I’ll briefly mention how I went about writing a paper.


In the book Science Fictions, Stuart Ritchie points out the reproducibility crisis in science. So many studies just don’t hold up when the experiments are re-run. Why? Because researchers are incentivised to publish papers. Not write good code, not verify other researchers’ work or re-run experiments. It’s all about how many papers you publish, and that’s just terrible!

Where to next?

Our program HOLLy is now out there in the world, ready to be used and verified by anyone who wants to have a go. I hope it’ll be useful for these working with high-throughput microscopy data. The story isn’t over yet. No doubt I’ll need to revise the code, update the documentation, maybe add some features or take away others. The paper itself is still under review at the moment — there will be corrections and things to change no doubt (It’s well understood that reviewer 2 is always a pain in the arse).



Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Benjamin Blundell

Benjamin Blundell


Freelance Research Software Engineer and Bioinformatics Student.