Machine understanding surgical actions from intervention procedure textbooks

Marco Bombieri, Marco Rospocher, Simone Paolo Ponzetto, Paolo Fiorini

Abstract: Highlights•Novel, Semantic Role Labeling-based approach for extracting procedural knowledge from surgical books and papers.•Comparison exploiting various language models in zero-shot, few-shot and full fine-tuning settings.•Public release of all the developed models.•Novel, combined metrics (predicate sense + argument role label) for assessing semantic role labeling performance.