Weakly supervised SNOMED labeling for Dutch radiology reports
by Koen Dercksen
In this talk, I will present some recent work on labeling Dutch radiology reports using weak supervision. First, we create a dataset using a combination of manual annotations and SNOMED ontology mapping. We iteratively update a feature based labeler with manual corrections to determine a threshold for diminishing returns. We also evaluate some machine learning labelers using the same technique, investigating which approach takes less manual labour.
-