Making Smarter Microscopes

Image by Konstantin Kolosov from Pixabay

I always joke with my students that the biggest difference between me and them when approaching a microscope is that there is nobody I can call for help with finding what I’m looking for. This might seem trivial, but I firmly believe that the microscope smells fear. When I walk up to it with confidence, it knows that I mean business. Contrast this with how the students approach that same microscope, with raging stress hormones, and I conclude that the microscope, like so many other things in life responds differently to confident people.

Okay, so of course this is somewhat sarcastic. The difference in confidence does make a difference, but more to how we approach setting up the microscope and scanning through the slide. As students get more experience, they gain confidence and become more skilled at finding things on a slide. But, it turns out, that smart microscopes may actually be on the horizon.

Muthumbi et al (2019) have combined two different elements (deep learning and sample illumination) to create a microscope that is able to accurately identify blood cells that are infected with malaria. With over 200 million cases worldwide in 2017 and almost half a million deaths (World Health Organization, n.d.), malaria is a significant health concern and finding a way to not only identify infections quickly and accurately is vital to treatment and management. But what makes this possible you might ask?

Muthumbi and their team set out to find out if they could combine machine learning with alternate lighting setups to increase the speed and accuracy of scanning blood samples for the malaria parasite. This is important because, while human identification is currently the most accurate it requires significant time. Imagine, the next time you are trying to find someone in a crowd that you can only look through the end of a toilet paper tube. It would take you a long time to scan through the entire crowd like this. This is what happens when you look through a microscope: although you can see small things because of the magnification, you can only look at a very small portion of the slide at a time, making this a very time consuming practice, on average about 10 minutes or more for experts. Now imagine that you could speed this up by having a computer scan each sample. Sounds helpful right? So this was the first question. But to answer it, the researchers had to solve an additional challenge. If you have ever fought with an office or housemate about how bright or dim a space should be you’ll be familiar with the second challenge: how should the slide be lit? Computers may need a different light setup, compared to humans, to best identify infected cells.

Malaria infection visible in a blood smear Photo: CDC

But, malaria infections generally occur in areas that are resource restricted, areas where we can’t rely on brand new, fancy microscopes. Therefore, researchers attached a set of LED lights that could be programmed to a regular light microscope. Then, they designed software that not only looked for the parasite but also adjusted the LED lights to achieve better contrast for that parasite identification. The microscope had to learn what it was dealing with and adjust the context to make it best able to complete the task. This achieved 5-10% greater accuracy depending on the type of blood smear used.

For most people living in North America this research is not really part of their everyday. But, it is relevant because it is an example of the social justice that we need to consider in science. We can’t just design experiments and assume that someone else will take care of what it means or how to apply it. In addition, there are questions that can only be triggered by the surrounding context. Muthumbi and the others on this project created different research questions by considering the social context that surrounded the issues of diagnosing malaria infections. They are addressing at least part of a social problem by conducting scientific research. That is finding the science in the everyday.

Muthumbi, A., Chaware, A., Kim, K., Zhou, K. C., Konda, P. C., Chen, R., Judkewitz, B., Erdmann, A, Kappes, B., & Horstmeyer, R. (2019). Learned sensing: jointly optimized microscope hardware for accurate image classification. Biomedical Optics Express 10, 6351-6369. https://doi.org/10.1364/BOE.10.006351

World Health Organization (n.d.) Malaria. https://www.who.int/malaria/en/