Literature · OpenMed
OpenMed DiseaseDetect
A biomedical transformer for detecting disease names and medical conditions in scientific and clinical text.
What OpenMed DiseaseDetect is
OpenMed DiseaseDetect (BioMed 335M) is a specialised token-classification model developed by OpenMed for biomedical named entity recognition (NER). Fine-tuned on disease-specific annotations, it is designed to locate mentions of diseases, syndromes, pathological states, and medical conditions within scientific abstracts, clinical records, and healthcare documents.
The model addresses the challenge of identifying varied medical terminology across unstructured biomedical text. By automating disease mention extraction, it aids researchers and informatics specialists working on tasks such as biocuration, adverse event surveillance, and the assembly of biomedical knowledge graphs from scientific literature.
How it works
The model is built upon the BiomedNLP-BiomedELECTRA-large-uncased-abstract backbone (containing approximately 335 million parameters) and fine-tuned for token classification using the Hugging Face Transformers framework. Optimisation is performed with the AdamW optimiser alongside learning rate scheduling.
Training and evaluation use the BC5CDR-Disease corpus, a disease-focused component derived from the BioCreative V Chemical-Disease Relation extraction dataset. This dataset comprises 1,500 PubMed abstracts containing 5,818 manually curated disease annotations. The model evaluates tokenised input text and assigns BIO-format entity tags (`B-DISEASE` and `I-DISEASE`), which mark the beginning and continuation of disease entities.
Key facts
- Base architecture
- BiomedNLP-BiomedELECTRA-large-uncased-abstract (ELECTRA / BERT-style transformer)
- Parameter count
- 335 million
- Task
- Token classification (Named entity recognition)
- Training dataset
- BC5CDR-Disease (1,500 PubMed abstracts, 5,818 annotated entities)
- Entity types
- B-DISEASE, I-DISEASE
- Primary language
- English
- Evaluation metrics (BC5CDR-Disease)
- F1 score 0.9005, Precision 0.8887, Recall 0.9126, Accuracy 0.9838
What you can use it for
- Biomedical literature mining
- Extract disease names, syndromes, and pathological conditions across PubMed abstracts and scientific articles for biocuration.
- Clinical text extraction
- Detect disease and condition references within clinical texts, patient documentation, and healthcare records.
- Knowledge graph construction
- Isolate disease entities from unstructured text to populate nodes in biomedical and drug discovery knowledge networks.
- Adverse event monitoring
- Identify mentions of emerging pathological conditions or secondary disorders in surveillance and clinical safety studies.
Run it here
No installation and no GPU of your own: fill in the inputs and it runs on our servers. Results stay in your browser unless you sign in.
Runs on On-demand GPU
Runs on our servers with on-demand compute. A first run needs time to load the model; active capacity can be reused and scales down when idle.Run this model
OpenMed DiseaseDetect
Strengths and limits
Good at
- Achieves a high recall of 0.9126 and an F1 score of 0.9005 on the standard BC5CDR-Disease benchmark.
- Built specifically on a domain-adapted biomedical ELECTRA backbone trained on scientific abstracts.
- Outputs standard BIO-tag annotations that map directly into standard Hugging Face token-classification aggregation pipelines.
Watch out for
- Identifies only disease entities (`B-DISEASE` and `I-DISEASE`) and cannot recognise other biomedical concepts like chemicals or genes.
- Trained exclusively on English biomedical literature and clinical text styles, limiting application to non-English sources.
- Ranks below three other OpenMed variants (SuperClinical-434M, PubMed-335M, and MultiMed-335M) in benchmark F1 score.
Inputs and outputs
You provide
Text
Entities are labelled token by token.
You get back
Labels with confidence scores, highest first.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use OpenMed DiseaseDetect commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
How do I run OpenMed DiseaseDetect without installing anything?
Use the form under “Run it here” on this page: fill in the inputs and press Run model. It runs on our servers on CPU only, typically in 12s, and a run costs about 28 credits. Visitors without an account get 1,000 credits a month.
What entity types can this model detect?
The model detects only disease entities, outputting `B-DISEASE` (the start of a disease term) and `I-DISEASE` (tokens continuing the entity).
What benchmark was used to evaluate this model?
It was evaluated on the BC5CDR-Disease corpus, where it achieved an F1 score of 0.9005, a precision of 0.8887, and a recall of 0.9126.
How are raw token predictions merged into complete entity spans?
Predictions can be grouped using token classification aggregation strategies such as 'simple', 'first', 'average', or 'max' to combine adjacent BIO-tagged tokens into unified disease names.
How to cite it
Panahi, M. (2025). OpenMed NER: Open-Source, Domain-Adapted State-of-the-Art Transformers for Biomedical NER Across 12 Public Datasets. arXiv:2508.01630 [cs.CL].
Written from OpenMed DiseaseDetect's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.