Computer Vision News - October 2022

60 Poster Presentation approaching real-world problems to make deep learning methods help clinically. “Professor Namburete is a great supervisor because she helps us to investigate the problems that we’re interested in but allows us to have our own research directions and creativity, ” she tells us. Nicola is not the only OMNI member showcasing their work at MICCAI this year. Linde Hesse’s poster on Monday, INSightR-Net: Interpretable Neural Network for Regression using Similarity- based Comparisons to Prototypical Examples , explored how to make networks interpretable for regression tasks. Also, Pak Hei Yeung ’s poster on Monday, Adaptive 3D Localization of 2D Freehand Ultrasound Brain Images , explored domain adaptation using cycle consistency for ultrasound. “I really like programming and computer vision, but I’m motivated by exploring real- world problems,” Nicola adds. “I like being able to identify barriers to current methods being applied clinically, and this was the first obvious barrier to approach.” features for the sites that we’re not currently training at to protect the privacy of the individuals.” Could this novel element be what led to the paper being accepted at MICCAI this year? “Yeah, I think it goes substantially beyond the existing approaches because they share the whole feature embedding,” she responds. “From that feature embedding, you would be able to reconstruct the image, especially if it was a segmentation task, so it doesn’t protect privacy.” The paper demonstrates this new framework for the task of age prediction. Going forward, Nicola hopes to generalize it to different architectures and tasks, such as segmentation. Perhaps that will earn her an oral presentation in Vancouver next year? “Hopefully!” she laughs. Nicola is currently based in a new group at Oxford called the OxfordMachine Learning in NeuroImaging Lab (OMNI) , working under Ana Namburete . Its work focuses on BEST OF MICCAI