The SMART Cumulus Text-to-FHIR NLP Pipeline
summary
This pipeline containerizes Apache cTAKES behind a Dockerized REST server, pulls notes via the FHIR Bulk Data API, maps cTAKES semantic types into FHIR resources (Condition/Observation/Procedure/MedicationStatement), and writes concept codes into resource.code.coding (medications into medication.codableConcept). It adds extensions for NLP metadata, versioning and an “nlp-derived” flag, uses an extended terminology that includes COVID-19 lists plus SNOMED/RxNorm, detects negation with rule-based negex (~80–90% on held sets), and targets scaling to millions of notes.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:29What is NLP
- 4:55Open Source Tools
- 7:40Project Overview
- 8:30Coverage Specific Terminology
- 10:00Docker Container
- 10:44Implementation
- 13:25Conclusion
- 13:59QA
- 17:09Negation
related talks
37mA novel way to manage and represent unstructured clinical data
41mInsights into patient health with
30mIntroduction to SNOMED with FHIR
41mMastering FHIR Terminology: Navigating advanced services
16mSoK: Automated TTP Extraction from CTI Reports – Are We There Yet?
19mFrom Data to Dollars: Improving Medical Billing Accuracy Using NLP
20mRough waters to smooth sailing: Records remodeling to FHIR
30mFrom Code to Clinic — Building an Internal Developer Platform for Clinical AI