AI-Based Software Development
for Standard Medical Terminology Mapping
Project Overview
A national research project developing AI-powered software to automatically map local clinical terms to international standard terminologies (SNOMED CT, ICD-10, LOINC, etc.), validated across 12 healthcare institutions.
-
Target Data: Diverse clinical elements under KR CDI — diagnoses, laboratory tests, procedures, medications, and more
-
Performance Target: Top-10 accuracy ≥ 90%, consistency ≥ 95%
Role of Our Research Team
The research team at WITH LAB leads three core workstreams:
① Mapping Strategy & Preprocessing Design Analyze composition and characteristics of collected data by KR CDI element
-
Select appropriate standard terminology per element (SNOMED CT, ICD-10, KCD, LOINC, RxNorm, etc.)
-
Investigate EMR data structures and input interfaces across institutions
-
Design item-level preprocessing strategies and determine mapping priorities
② Ground Truth (GT) Dataset Construction
-
Form a specialized mapping team with clinical experience and SNOMED CT/LOINC expertise
-
4-step validation: Independent mapping → Internal consensus → External expert review → Finalized GT
-
Target: 3,000 GT entries by Year 2; additional 3,000 by Year 4
③ AI Software Performance & Usability Evaluation
-
Mapping quality evaluation using Exact / Broad / Narrow / Wrong / NotMap framework
-
Performance evaluation: Top-10 accuracy and consistency measurement
-
Usability evaluation: Heuristic evaluation and System Usability Scale (SUS) with medical informatics experts
-
Field validation support and UI/UX improvement recommendations
Annual Milestones
Mappers
