Elisabeth Wetzer

I am a PhD student in Computerised Image Processing at the Division for Vi3 at the Dept. of Information Technology.
I develop methods in image analysis with a focus on texture representation in machine learning for biomedical applications and belong to the Methods in Image Data Analysis group.

I am a graduate student at the Centre for Interdisciplinary Mathematics (CIM), for which I am PhD representative for the Faculty of Science and Technology on the board.

Academia
I have received my Bachelor of Science in Technical Mathematics at the TU Wien in 2014 and my Masters of Science in Biomedical Engineering in 2017 at TU Wien. I spent a year on exchange through Erasmus at Umeå University studying Computational Physics.

I was a visiting research student at the University of Southern California at the Nuzhdin Research Lab in 2013 working on tracking algorithms in behavioural studies of Drosophila melanogaster.

In 2017 I was a research intern at the National Institute of Informatics (NII) in Tokyo, Japan in Michael E. Houle's group, working on estimates of intrinsic dimensionality of data.

In spring 2019 I've participated in the thematic trimester at the Institut Henri Poincaré in Paris on The Mathematics of Imaging.

I am a member of the association of European Women in Mathematics, the Uppsala University student chapter of the Society for Industrial and Applied Mathematics (SIAM) and I am on the editorial board of the biannual newsletter of the Swedish Society for Automated Image Analysis (SSBA), which is available on the SSBA website and informs of the current status of research and events related to image analysis throughout Sweden.

Public Outreach
I was a speaker at the Soapbox Science event in Uppsala on May 25, 2019 and at the Pint of Science on May 20, 2019.

Representation Learning for Multimodal Images
Images acquired in multiple modalities can provide complimentary information about a sample compared to a single modality. Often these images have to be acquired in different machines and hence require image alignment prior to their fusion. Multiomodal image registration is a very challenging task, in particular if the modalities differ strongly in their appearance and the images in the respective modalities only share few common structures. We hence adopt contrastive learning regimes to generate image representations for multimodal image pairs which are similar with respect to a chosen similarity measure and can be used in combination with monomodal registration methods to achieve alignment of multimodal image pairs.

Multi-layer object representations for texture analysis
Texture features such as local binary patterns have shown to provide complementary information to that of plain intensity images in learning algorithms. We investigate methods on the fusion of texture and intensity sources, as well as the problems connected to the fact that many texture descriptors are unordered sets and require suitable (dis-)similarity measures in order to be processes by for example convolutional neural networks. We develop strategies to integrate more complex texture features into learning methods and evaluate their performance on various biomedical images. Such hybrid object representations show promising results in detection and segmentation of high resolution transmission electron microscope (TEM) images, taking it one step closer to automation of pathological diagnostics.