From BERT to generative AI - Comparing encoder-only vs. large language models in a cohort of lung cancer patients for named entity recognition in unstructured medical reports

Kamyar Arzideh, Henning Schäfer, Héctor Allende-Cid, Giulia Baldini, Thomas Hilser, Ahmad Idrissi-Yaghir, Katharina Laue, Nilesh Chakraborty, Niclas Doll, Dario Antweiler, Katrin Klug, Niklas Beck, Sven Giesselbach, Christoph M. Friedrich, Felix Nensa, Martin Schuler, René Hosch

Abstract: Highlights•Study analyzed 2013 real-world pathology and 413 radiology reports for NER training.•Evaluation included comparison of encoder and decoder language models.•Encoder NER models outperform LLMs in extracting entities from medical reports.•LLMs show high precision but poor recall in clinical entity extraction.

External IDs:dblp:journals/cbm/ArzidehSABHILCDAKBGFNSH25