Clinical Information Extraction
Overview
Clinical information extraction converts free-text notes, radiology and pathology reports, trial documents, and patient narratives into structured entities, attributes, relations, codes, and timelines for cohort search, surveillance, and decision support. Systems range from rules and CRF- or BERT-style sequence taggers to biomedical transformers and prompted or fine-tuned large language models, with retrieval-augmented, ontology-aware, and neural-symbolic pipelines supporting normalization, temporal linking, and validation. Evaluation uses span and entity F1, relation and field accuracy on resources such as MIMIC and radiology-report benchmarks, and increasingly tests evidence grounding, cross-site transfer, multilingual performance, and downstream utility. Large language models handle flexible schemas, but reliability remains uneven across domains and annotation conventions, especially for negation, uncertainty, missing data, and calibration; evidence-linked review and locally deployable models help address these risks.
Latest papers
106 papers11 papers in the last four weeks, down 15% on the four weeks before. 0.1% of all new papers.