At the 40th International Conference on Computer Assisted Radiology and Surgery (CARS 2025) in Berlin, Raphaela Märkl, a master's student in computer science with a research option, presented her work
‘Enhancing Generalisation in Zero-Shot Multi-Label Endoscopic Instrument Classification’
as part of the ISCAS (International Society for Computer Aided Surgery) sub-event in the main auditorium – an important milestone for a young scientist.
The work investigates how AI models can recognise surgical instruments in endoscopic images without having seen them before in training (zero-shot learning). Through the use of semantically rich text inputs generated with ChatGPT, Sentence-BERT and z-score normalisation, recognition performance was significantly improved, especially for unknown instrument classes.
The results were published in advance in the International Journal of Computer Assisted Radiology and Surgery (IJCARS) by Springer Nature.
The work was carried out in the ReMIC laboratory at OTH Regensburg under the supervision of Prof. Dr. Christoph Palm with co-authors Tobias Rückert, David Rauber, Max Gutbrod and Danilo Weber Nunes. The presentation and publication underscore the high quality of research in the master's programme at OTH Regensburg.
