39 Computer Vision News Computer Vision News SynthBrainGrow UKRAINE CORNER When Anna initially had the idea for this work, she had difficulty finding structurally correct, anatomically correct, synthetically generated brains. “Usually, when people generate synthetic imaging, they do it in 2D because of the capacity constraint,” she explains. “It’s a really computationally extensive task. However, as soon as you concatenate 2D images into a 3D volume, they don’t really make sense because of the slight variance between slices. When you try to combine a 3D volume and then perform downstream tasks like segmentation, it doesn’t work. It becomes like slices from an apple and a peach. Different fruits together just don’t make the same object anymore.” To solve this, Anna developed her method in 3D, ensuring downstream tasks could still run on these synthetically generated brains. Anna presented her work at the Deep Generative Models workshop at MICCAI (DGM4MICCAI) earlier this month. Unfortunately, she could not attend Marrakesh in person, but the workshop organizers allowed her to record a video presentation. “I’ve been a huge fan of this workshop,” she shares. “I attended last year, and it was brilliant. There was such a great collaboration and exchange of ideas. Diffusion model is a very young
RkJQdWJsaXNoZXIy NTc3NzU=