Application of Vision Foundation Models in Computational Pathology

Dates & Times: 19th, 20th, 21st May 2026, Time: 9:00-12:00 each day

Location: Seminar room 0.04, Hamburg Center for Translational Immunology (HCTI), Building N25, UKE

Instructor: Dr. Patrick Fuhlert, Institute of Medical Systems Bioinformatics, UKE

Language: English

Prerequisites: Experience working in Python, model development with pytorch, working with git.

Description: This workshop provides a hands-on introduction to applying vision foundation models (VFMs) in digital pathology, with techniques that are broadly applicable across other image-based domains. VFMs are large-scale vision architectures pre-trained on diverse, high-volume whole slide image corpora to produce reusable visual representations that capture broadly applicable semantic and structural information across image domains. In digital pathology, VFMs are trained on massive collections of histopathology image patches, claiming robust, domain-specific feature extraction. In this workshop, we utilize fixed pre-trained embeddings to study downstream prediction tasks, explicitly focusing on the representational capacity of foundation models. The emphasis lies on keeping model development cycles short and on producing meaningful results on relevant downstream tasks in digital pathology.

Topics:

• Vision foundation models for digital pathology: overview of pretrained VFM architectures (with a focus on UNI2) and their representational properties in histopathology.
• Fixed-feature learning and embedding-based pipelines: use of frozen VFM embeddings for downstream prediction, avoiding end-to-end retraining while enabling rapid experimentation.
• Downstream evaluation: practical application of VFM embeddings to digital pathology tasks, with emphasis on reproducibility and efficient model development cycles.

Registration for this workshop will open soon