Mind the Gap: Scanner-induced domain shifts pose challenges for representation learning in histopathology [ISBI23 paper]

Frauke Wilm from our group has presented her latest work on overcoming domain shifts in histopathology on the 2023 IEEE International Symposium on Biomedical Imaging. In brief, our innovative method, named Barlow Triplets, was designed to create scanner-agnostic image representations, utilizing a dataset procured from multiple scanners. Though our method effectively synchronized different scanner outputs, it only brought about minor enhancements in our main focus of tumor segmentation. This research underscores the significant impact of scanner characteristics on histopathology tasks. It also brings new insights into why the success of self-supervised techniques, though well-documented in the realm of natural images, hasn’t been as prominent in the field of histopathology.

Find the paper here: [preprint]