Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

PSMA-Targeted Radioligand Therapy Beyond the Post-Taxane Setting: A Review of Evidence Across the Prostate Cancer Spectrum.

Cancers·2026
Same author

Integrating Clinical Classifications Software Refined, Process Indicators, and Geographic Information System Mapping to Inform Population Health Management: Development of an Interactive Dashboard.

JMIR medical informatics·2026
Same author

Impact of an Ambient AI Scribe Among Clinicians and Patients: Real-World Prospective Observational Time-Motion Study.

JMIR medical informatics·2026
Same author

DILIConsult: A Multi-Agent Large Language Model Framework for Evaluating Drug-Induced Liver Injury in ICU Settings.

Pharmacotherapy·2026
Same author

Ensemble Machine Learning Models for Predicting Patients With High Usage: Model Validation and Economic Impact Analysis.

JMIR medical informatics·2026
Same author

Cost-effectiveness analysis of a biopsy-free diagnostic strategy for prostate cancer using mpMRI and PSMA-PET/CT.

BJUI compass·2026

Related Experiment Video

Updated: Sep 10, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

2.0K

A Multimodal Large Language Model as an End-to-End Classifier of Thyroid Nodule Malignancy Risk: Usability Study.

Gerald Gui Ren Sng1,2, Yi Xiang3, Daniel Yan Zheng Lim2,4

  • 1Department of Endocrinology, Singapore General Hospital, 20 College Road, Academia Level 3, Singapore, 169856, Singapore, 65 63214377.

JMIR Formative Research
|August 19, 2025
PubMed
Summary

Multimodal large language models (LLMs) show potential for thyroid nodule risk stratification, but accuracy remains suboptimal for clinical use. Commercial models offer better consistency, while prompt engineering improves reliability.

Keywords:
artificial intelligencelarge language modelsmultimodalrisk stratification systemsthyroid nodules

More Related Videos

Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
07:13

Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model

Published on: April 18, 2025

193
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

681

Related Experiment Videos

Last Updated: Sep 10, 2025

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
04:23

A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

Published on: April 21, 2023

2.0K
Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model
07:13

Comparison of Predictive Performance of Three Lymph Node Staging Systems in Colorectal Signet Ring Cell Carcinoma Based on Machine Learning Model

Published on: April 18, 2025

193
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
03:14

Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness

Published on: December 6, 2024

681

Area of Science:

  • Artificial Intelligence in Medical Imaging
  • Multimodal Large Language Models (LLMs)
  • Thyroid Nodule Diagnostics

Background:

  • Thyroid nodules are commonly assessed via ultrasound imaging.
  • Current risk stratification systems (e.g., ACR TI-RADS) face challenges with interobserver variability and low specificity.
  • Multimodal LLMs offer potential for improving diagnostic efficiency but require validation.

Purpose of the Study:

  • To evaluate the accuracy and consistency of multimodal LLMs for thyroid nodule risk stratification using the ACR TI-RADS system.
  • To assess the impact of fine-tuning, image annotation, and prompt engineering on LLM performance.
  • To compare the performance of open-source versus commercial multimodal LLMs.

Main Methods:

  • Three multimodal LLMs were evaluated: LLaVA, LLaVA-Med, and OpenAI's o3 model.
  • 192 thyroid nodules from public ultrasound datasets were assessed.
  • Models were tested with basic and modified prompts, and with unlabeled versus annotated images, generating 6912 responses.

Main Results:

  • The commercial o3 model showed the highest validity (98.6%) and accuracy (up to 57.3%), though still suboptimal.
  • Prompt engineering improved accuracy for composition but reduced it for shape, margins, and overall classification.
  • Image labeling marginally improved accuracy for nodule margins with LLaVA models; consistency was highest with o3 and improved with labeling/modified prompts.

Conclusions:

  • Multimodal LLMs show promise but require further development for reliable clinical integration in thyroid imaging.
  • Commercial models currently outperform open-source options in accuracy and consistency.
  • Prompt engineering significantly enhances consistency, particularly in commercial models, highlighting its importance for optimization.