Computer Vision News - November 2020

3 Nils Eckstein 27 Best of MICCAI 2020 for automatic neuron segmentation still make mistakes and even a one per cent error rate can give the wrong connectome. Microtubules directly follow the morphology of neurons and so can help by providing a second information channel for proof-reading. “ Automating proof-reading of neuron segmentation is a big problem,” Nils points out. “People spend millions of dollars to do this manually right now. If you can fix that, it is huge.” Deep learning isn’t enough alone to solvethisproblemalone, sohecombines it with the more classical technique of constraint optimization using integer linear programming to speed it up. “The marriage of these two things was very interesting,” he reveals. “Everybody praises deep learning, and it is amazing, but it is not the end-all solution to every problem .” Nils says the work has already been used in various projects and has many practical applications, including segmenting all kinds of organisms and cells. Their work to speed up the optimization problem is particularly elegant and efficient, which he imagines could be used for other tracking problems that have the same form and constraints, with structures that are long and do not branch. Finally, we ask, how does he feel to have been acknowledged with a prestigious Young Scientist award ? “It feels so good to get this recognition,” he smiles. “Going forward in my research career, I hope it will open up other opportunities and further possibilities for grants. That would be nice!”