Related Experiment Video
Updated: Jun 4, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
A Hybrid Rule-Based and Large Language Model Artificial Intelligence System for Electrodiagnostic Reporting:
Yesung Jung1, Min Kyun Sohn1,2, Ja Young Choi1
1Department of Rehabilitation Medicine, Chungnam National University Hospital, Daejeon, Korea.
Muscle & Nerve
|June 2, 2026
Summary
A new system using a large language model (LLM) successfully drafts electrodiagnostic (EDX) reports from nerve conduction studies (NCS) and electromyography (EMG) measurements, showing high agreement with clinician interpretations.
Area of Science:
- Medical Informatics
- Neurology
- Artificial Intelligence
Background:
- Electrodiagnostic (EDX) studies, including nerve conduction studies (NCS) and needle electromyography (EMG), are crucial for diagnosing neurological conditions.
- Current EDX reporting practices are often inconsistent across laboratories and labor-intensive, leading to documentation challenges.
Purpose of the Study:
- To evaluate a novel hybrid system combining rule-based logic and a constrained large language model (LLM) for automated drafting of EDX reports.
- To assess the system's ability to generate standardized EDX reports from structured EDX measurements.
Main Methods:
- A retrospective analysis of 332 EDX reports from two institutions was performed.
- A rule-based system extracted standardized abnormal findings, and a label-constrained LLM drafted diagnostic interpretations from a predefined inventory.
- Agreement with clinician interpretations was measured using the Coverage-Localization Index (CLI).
Main Results:
- The system achieved high overall agreement with clinician diagnostic interpretations (mean total CLI 96.7%).
- The rule-based component accurately captured all defined abnormal findings.
- Agreement was slightly lower for more complex examinations.
Conclusions:
- The hybrid LLM system effectively drafts standardized EDX reports with high concordance to clinician diagnoses.
- The system demonstrates potential for improving efficiency and consistency in EDX reporting.
- Further validation on unselected, routine EDX reports, including normal and ambiguous cases, is warranted to assess real-world performance and workflow impact.