Inspiration
Diatoms are tiny algae with intricate glass shells, and labs study them under electron microscopes. But measuring them is slow, manual work: someone has to trace each shell, count hundreds of tiny pores, and decide by eye whether it's broken. A single image can take a long time, and two people often get different answers. We wanted to give researchers that time back and make the results consistent.
What it does
You drop in microscope images of diatoms, and the app does the rest. It works out the image scale on its own, outlines every diatom, measures its size and shape, counts and sizes its pores, and flags whether each one is intact, cracked, or broken. If it gets something wrong, you can fix it with a click. Results come out as labelled images and a spreadsheet, and everything runs on your own computer, so no data leaves the lab.
How we built it
We built it in Python with a simple point-and-click web interface. It uses image-processing techniques to find the diatoms and their pores, and reads the scale straight from the information the microscope saves with each image. We tested it on real images from our lab, including Didymo, cultured Thalassiosira, and field samples.
Challenges we ran into
Real microscope images are messy. Some samples are crowded with hundreds of overlapping diatoms, others are full of debris that looks a lot like broken shells. Early versions mistook natural features of healthy diatoms, like the slit that runs down the middle and the narrow "waists" of some species, for cracks. Every microscope brand also stores its scale differently, so we had to teach the app to read several formats.
Accomplishments that we're proud of
It reads the scale automatically from five different microscope brands, so users never have to measure a scale bar by hand. It can tell natural shell shapes apart from real damage. And it's built for people who aren't programmers: anyone in the lab can open it, drop in images, and get a spreadsheet.
What we learned
Real lab data is much harder than clean examples. We learned how much a diatom expert knows without thinking about it, like what counts as "broken", and how hard it is to turn that into rules a computer can follow. We also learned that letting users correct mistakes matters more than trying to be perfect.
What's next for Diatom SEM Analyzer
Next, we'll use the corrections lab members make in the app to train an AI model on our own images, so it keeps getting better at crowded samples. We also want to grow a reference library so it can identify species automatically, and share it with other labs that study diatoms.
Log in or sign up for Devpost to join the conversation.