Computer Vision News - February 2023

45 RnR-ExM Robust Non-rigid Registration... why precise registration is essential. ” Donglai Wei completed his PhD at MIT and his postdoc at Harvard, which is how he linked up with Emma and Ruihan to work on this project. He has been contacted by people from different labs excited about this competition and hoping it will become an annual event so they can participate and donate their data in the future. They view it as an opportunity to create a bigger data set and a more challenging problem. “ Registration is a classic problem for microscopy, but we all want a deep learning revolution for this field, ” he tells us. “ The problem is a lack of data, but ImageNet was not built in a day . We only have a small data set, but we hope to make it more complete and diverse over the years. This project is only one small step towards making expansion microscopy a more sure technology – not only imaging wise but also processing wise to support neuroscientists’ investigations. ” Ruihan adds: “ If we solve this computational problem, more wet lab biologists will use our technology because they’ll see how easy it is. It’s an iterative process. If we make the accuracy 80%, then labs that can generate high-quality data sets can already use it. If we make it 90% accurate, people who prepare less good samples can use it. If we lower it, then everybody can use it. There’s always an audience for our method. ” The team want this challenge to facilitate a discussion around expansion microscopy registration and encourage the release of more open-source code for registering this type of data. Many labs have developed in- house methods, but the hope is that this will create something more communal. Emma will be graduating in the spring. Meanwhile, she is working full-time on her thesis, of which this challenge has been a big part. “ I felt confident that I could do it, even though it would be a lot of work, ” she recalls. “ The results of this challenge will give important information to the Boyden lab and the Visual Computing Group at Harvard , two groups that I’m affiliated with. The kind of infrastructure we’ve set up, the evaluation tools, the website, everything can be used for future iterations of this when we’re building up our microscopy dataset. I wanted to be a part of something that would have useful results for the lab but also create a long- term impact. ”