Talk by Prof. Marc Niethammer

Abstract: Towards more general approaches for medical image registration

Image registration, the process of establishing spatial correspondences between images, is a key task in medical image computing. For example, to facilitate analyses in a common atlas spatial coordinate system or to track changes over time in radiation treatment planning. Image registration approaches have been extensively developed over the previous decades with a shift to approaches based on deep learning over the last decade. However, while modern deep learning registration approaches allow for highly accurate and fast registrations many existing approaches are task-specific. Hence, these approaches require extensive retraining or fine tuning for a new registration task. This talk will provide an overview of approaches to obtain deep registration networks that generalize across image types and registration tasks. It will also discuss some recent work on building deep registration models with desirable equivariance properties as a further step towards registration models that work well in practice and that behave somewhat more predictably.

Bio

Marc Niethammer is a professor in Computer Science and Engineering with a joint appointment in Neurological Surgery at the University of California, San Diego (UCSD). He holds the Halıcıoğlu Endowed Chair in Health AI and leads the Biomedical Image Analysis Group. Before joining UCSD, he was a Professor of Computer Science at the University of North Carolina at Chapel Hill (UNC) from 2008 to 2024. Dr. Niethammer was the program chair for the International Conference on Information Processing in Medical Imaging (IPMI, 2017) and is an Executive Editor of MELBA, the Journal for Machine Learning for Biomedical Imaging. Dr. Niethammer’s work focuses on methods for statistical shape analysis, image segmentation, image registration, machine learning, and related applications.