Jump to main content

Quicklinks

Institution logo

Prof. Dr. Katarzyna Bozek

University of Cologne
Center for Molecular Medicine Cologne
Robert-Koch-Str. 21
50931 Cologne
Email: k.bozek(at)uni-koeln.de
URL: bozeklab.com
ORCID: 0000-0002-0917-6876

CDS Research Areas

CDS Selected Projects

Our interests lie in development of deep learning methods for extraction of quantitative information from image and time series data in biology and medicine. Via supervised and self-supervised approaches we aim to build meaningful representations of such data that allow for their further downstream analysis e.g. in prediction of clinical outcomes, discovery of patterns in morphology and molecular structure organization. Our research is applied in several branches of biomedical research – in pathology, nephrology, eye medicine, and neurology.

Representation space of behavior of two interacting mice. Each dot corresponds to motion data of the mice in a time span of 1 sec. The representations were obtained via a Transformer neural network model trained in a fully self-supervised manner. Even though no action labels were used in model training, there is a clear clustering of behaviors in the representation space corresponding to their action labels.

Selected publications

  1. Pisula JI, Bozek K, Fine-Tuning a Multiple Instance Learning Feature Extractor with Masked Context Modelling and Knowledge Distillation. (2025) European Conference for Computer Vision – ECCV 2024 Workshops. Lecture Notes in Computer Science, vol 15638. Springer, https://doi.org/10.1007/978-3-031-91721-9_1
  2. Rose F, Michaluk M, Blindauer T, Ignatowska-Jankowska BM, O’Shaughnessy L, Stephens GJ, Pereira TD, Uusisaari MY, Bozek K, Deep Imputation for Skeleton Data (DISK) for Behavioral Science (2025), Nature Methods, https://doi.org/10.1101/2024.05.03.592173,
  3. Pisula JI, Helbig D, Sancéré L, Persa O-D, Bürger C, Fröhlich A, Lorenz C, Bingmann S, Niebel D, Drexler K, Landsberg J, Thomas R, Bozek K*, Brägelmann J*, (2025) Explainable, federated deep learning model predicts disease progression risk of cutaneous squamous cell carcinoma Nature (2025) njp Precision Oncology, 9, 205. https://doi.org/10.1038/s41698-025-00997-4
  4. Deserno M.,  Bozek K., Unsupervised Representation Learning of C. elegans Poses and Behavior Sequences From Microscope Video Recordings (2025) eLife14:RP106593 doi: 10.7554/eLife.106593.1
  5. Butt L, Unnersjö-Jess D, Höhne M, Sergei G, Witasp A, Wernerson A, Patrakka J, Hoyer PF, Blom H, Schermer B, Bozek K*, Benzing T*, Deep learning-based segmentation and quantification of podocyte foot process morphology Kidney International (2023), https://doi.org/10.1016/j.kint.2023.03.013