Related Experiment Video
Updated: Aug 11, 2026
06:28
Screening People on Standing Balance with Romberg Testing and Walking Balance with Tandem Walking
Published on: September 1, 2023
Conversational Large Language Models for Vestibular Diagnosis in Outpatient Clinics: Prospective Multicenter
Chongkai Lu1, Ruiqi Zhang1, Huaili Jiang2
1ENT Institute and Department of Otorhinolaryngology, Eye and ENT Hospital, Fudan University, 83 Fenyang Road, Shanghai, Shanghai, 200031, China, 86 13524844652.
Journal of Medical Internet Research
|August 10, 2026
Summary
A conversational Large Language Model (LLM) agent achieved nearly 80% diagnostic accuracy for vestibular disorders in a clinical trial. This tool shows promise for improving diagnosis in settings with limited specialist access.
Area of Science:
- Medical Informatics
- Artificial Intelligence in Healthcare
- Otolaryngology
Background:
- Vestibular disorders are common, often misdiagnosed due to incomplete or unstructured history-taking in non-specialist settings.
- Existing digital health tools for symptom elicitation typically use static questionnaires or rule-based logic.
- Large Language Models (LLMs) offer a flexible, adaptive natural-language approach, but clinical evidence is limited.
Purpose of the Study:
- To benchmark the diagnostic performance of various LLMs using static vestibular histories.
- To prospectively evaluate a locally deployed conversational LLM agent within outpatient vertigo clinics.
Main Methods:
- A two-phase diagnostic accuracy study was conducted.
- Phase 1 (History-Based Evaluation): 10 LLMs and 5 otolaryngologists reviewed 227 structured vertigo histories.
- Phase 2 (Prospective Clinical Evaluation): A nurse-assisted, tablet-based LLM agent (DeepSeek-R1) conducted natural-language dialogues with 176 outpatients across 5 centers; the agent only received history information.
Main Results:
- In the History-Based Evaluation, top LLMs achieved 69.6% Top-1 accuracy, not significantly outperforming the specialist panel (63.9%).
- In the Prospective Clinical Evaluation, the conversational agent achieved 79.55% concordance with reference diagnoses.
- Concordance was highest for benign paroxysmal positional vertigo (97.73%) and Ménière disease (90.00%), lower for vestibular migraine (71.11%) and persistent postural-perceptual dizziness (67.86%).
Conclusions:
- A locally deployed, nurse-assisted conversational LLM agent demonstrated approximately 80% diagnostic concordance in a prospective multicenter study.
- The LLM agent showed particularly high performance for benign paroxysmal positional vertigo.
- Findings support LLMs as clinician-facing tools for structured history-taking and diagnostic support, especially where vestibular expertise is limited.
Related Concept Videos
01:29The Vestibular System
The vestibular system is a set of inner ear structures that provide a sense of balance and spatial orientation. This system is comprised of structures within the labyrinth of the inner ear, including the cochlea and two otolith organs—the utricle and saccule. The labyrinth also contains three semicircular canals—superior, posterior, and horizontal—that are oriented on different planes.
01:15Equilibrium and Balance
The inner ear assumes dual functionalities of auditory perception and equilibrium maintenance. The vestibule is the organ responsible for balance. This organ contains mechanoreceptors, specifically hair cells, endowed with stereocilia, which aid in deciphering information regarding the position and motion of our heads. Two intrinsic components, the utricle and saccule, help perceive head position, while the semicircular canals track head movement. Neurological messages initiated in the...