Nataša Sladoje: Automated alignment of multimodal images: To (deep) learn – or not?

Multimodal imaging gives an opportunity to collect diverse and complementary information about a specimen, enabling a deeper understanding of complex interactions of the specimen’s biological, chemical, dynamic, and other properties. For successful correlation and fusion of information present in such diverse images, the images need to be accurately aligned. Automated multimodal image registration is a very challenging task, nowadays typically addressed by learning-based approaches. However, iterative (non-learning based approaches) seem to still stand the competition.

This talk will briefly introduce the topic of multimodal image registration, main challenges, some recent multimodal (bio)image registration methods, and will discuss advantages and disadvantages of learning- and non-learning registration approaches, with an aim to support selection of a suitable method for a given context, and to point to directions for further development of the field.

Nataša Sladoje is professor in computerised image analysis at the Department of Information Technology, Uppsala University. She is the scientific leader of MIDA - Methods for Image Data Analysis group, and the director of the Centre for Image Analysis (Uppsala University). She is the initiator and coordinator of the Master’s programme in Image Analysis and Machine Learning at the IT Department at Uppsala University.

Nataša obtained her PhD degree in computerized image processing at the Centre for Image Analysis, SLU, in 2005. Her scientific interests are in the development of methods for robust image processing with high information preservation, as well as the development and application of machine and deep learning methods for image processing and analysis, mostly driven by applications within medicine and bio-medicine. Her research has been funded by, among others, EU Horison 2020, Swedish Research Council, VINNOVA, and Cancerfonden (Swedish Cancer Society). Her recent work is particularly focused on robust image registration approaches.