MedCAT

MedCAT (Medical Concept Annotation Tool) represents the pinnacle of open-source clinical Natural Language Processing (NLP) architecture, specifically engineered for Named Entity Recognition and Linking (NER+L). Developed by the CogStack team at King's College London, it addresses the massive challenge of unstructured data in Electronic Health Records (EHRs). By 2026, MedCAT has matured into a foundational layer for clinical data science, utilizing a unique combination of shallow neural networks and unsupervised learning for concept extraction, followed by supervised online learning for context-aware disambiguation. Unlike traditional rule-based systems, MedCAT learns medical semantics directly from large-scale corpora, allowing it to map text to standardized ontologies such as SNOMED CT, ICD-10, and UMLS with high precision. Its architecture is built atop spaCy, ensuring high-throughput performance capable of processing millions of patient records. The framework's 'Learning while Annotating' capability allows clinical experts to refine models in real-time, drastically reducing the cold-start problem in medical entity extraction. As healthcare moves toward the OMOP Common Data Model (CDM), MedCAT serves as the critical translation engine between raw clinician notes and structured research databases.