Computer Vision News - November 2022

50 Congrats, Doctor!! Alessa Hering has recently completed her PhD at RadboudUMC Nijmegen. Her research focused on the development of deep-learning-based image registration methods and tumor follow-up analysis (link) . During her PhD, she mainly worked at Fraunhofer MEVIS. Currently, she pursues her career as a PostDoc at RadboudUMC but she still holds a position at Fraunhofer MEVIS. Congrats, Doctor Alessa! Deep-Learning-Based Image Registration: Medical image registration aims to align the anatomical structures of two images by establishing spatial correspondences. This is an important step for many tasks in medical image analysis as it links previously unrelateddata and enables joint processing of these data. While deep learning has become a methodology of choice in many areas like segmentation and classification, image registration is often still based on conventional methods. This is because we cannot ask a medical expert to annotate a reference deformation field that establishes such correspondences for every voxel of the images. In our research, we overcome this issue by using several loss functions that integrate prior anatomical knowledge as visualized in Figure 1. Furthermore, we developed a multi-level approach for deep-learning-based approaches to better handle deformations on different scales. The experimental results show very good performance on several publicly available datasets. Andasoneopponentcorrectlypointedoutduring my defense: you cannot seem to finish your PhD in our group without organizing a challenge - so I co-organized the Learn2Reg registration challenge mainly together withMattias Heinrich and Lasse Hansen to compare deep-learning- based to conventional registration methods on several registration tasks.