Table 1 Comparison of GatorTron with existing biomedical and clinical transformer models for clinical concept extraction and medical relation extraction.

Transformer	Clinical concept extraction									Medical relation extraction
	2010 i2b2³⁹			2012 i2b2⁴⁰			2018 n2c2⁴¹			2018 n2c2⁴¹
	Precision	Recall	F1 score	Precision	Recall	F1 score	Precision	Recall	F1 score	Precision	Recall	F1 score
BioBERT	0.8693	0.8653	0.8673	0.7478	0.8037	0.7747	0.8634	0.8921	0.8775	0.9663	0.9451	0.9555
ClinicalBERT	NA	NA	0.8780	NA	NA	0.7890	0.8592	0.8832	0.8710	0.9678	0.9414	0.9544
BioMegatron	0.8614	0.8761	0.8687	0.7591	0.8031	0.7805	0.8707	0.8915	0.8810	0.9711	0.9434	0.9571
GatorTron-base (1/4 data)	0.8682	0.9046	0.8860	0.7514	0.8013	0.7755	0.8772	0.8992	0.8881	0.9724	0.9457	0.9589
GatorTron-base	0.8748	0.9043	0.8893	0.7644	0.8221	0.7922	0.8759	0.9038	0.8896	0.9719	0.9482	0.9599
GatorTron-medium	0.8869	0.9122	0.8994	0.7812	0.8245	0.8022	0.8954	0.9035	0.8994	0.9721	0.9503	0.9611
GatorTron-large	0.8880	0.9116	0.8996	0.7862	0.8333	0.8091	0.8979	0.9021	0.9000	0.9776	0.9482	0.9627

Clinical concepts in 2010 i2b2 and 2012 i2b2 challenges: problems, treatments, lab tests; clinical concepts in 2018 n2c2 challenge: drugs, adverse events, and drug-related attributes (e.g., dose). Medical relation in 2018 n2c2 challenge: drug induced adverse events. Best F1 scores are presented in bold. NA: scores not reported.

Quick links

Search