Clinical Semantic Enrichment with Calibration and FHIR Interoperability
کلمات کلیدی:
Clinical NLP, semantic enrichment, entity linking, SNOMED CT, LOINC, UMLS, FHIR, calibration, de-identification, interoperabilityچکیده
Clinical notes and reports contain high-value signals for analytics and care pathways, but they are heterogeneous, noisy, and riddled with privacy-sensitive details. Building on the bibliometric evidence base in [20], we design an applied pipeline for clinical semantic enrichment that (i) performs PHI redaction, (ii) recognizes entities and links them to SNOMED CT and LOINC through UMLS, (iii) calibrates cross-encoder scores for risk-aware operation, and (iv) exports interoperable resources in HL7 FHIR. Evaluated on mixed corpora (discharge summaries, radiology reports, lab narratives), the approach improves candidate PR and macro-F1 while reducing latency. We provide reproducible figures (architecture, PR curves, reliability, latency), two tables (metrics and ontology coverage), and deployment guidance for hospital IT and research teams.
