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

Weitiao No. 3 increased isobutyric acid to enhance the effect of PD-L1 inhibitor on tumor growth and anti-tumor effect of CD8<sup>+</sup>T cells in gastric cancer.

3 Biotech·2026
Same author

Improving preoperative risk stratification in colorectal liver metastases: a multi-institutional evaluation of multimodal prediction models.

Scientific reports·2026
Same author

The impact of different early rehabilitation initiation times on functional recovery after flexor tendon repair.

Medicine·2026
Same author

WD-3 improves anti-PD-L1 therapy by remodeling the tumor immune microenvironment through gut microbiota.

Immunobiology·2026
Same author

Radiomic Carotid Plaque Features Integrated into Machine Learning Models for Cardiovascular Risk Prediction.

Ultrasound in medicine & biology·2026
Same author

Pancreas Cancer Risk During Intraductal Papillary Mucinous Neoplasm Surveillance - A SEER-Based Comparison.

Annals of surgery·2026

Related Experiment Video

Updated: May 27, 2025

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
07:15

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model

Published on: August 16, 2020

6.6K

Targeted generative data augmentation for automatic metastases detection from free-text radiology reports.

Maede Ashofteh Barabadi1, Xiaodan Zhu1, Wai Yip Chan1

  • 1Ingenuity Labs Research Institute, Department of Electrical and Computer Engineering, Queen's University, Kingston, ON, Canada.

Frontiers in Artificial Intelligence
|February 21, 2025
PubMed
Summary

Large language models like Llama3 can generate synthetic data to improve automated cancer metastasis detection from radiology reports. This approach enhances diagnostic accuracy using limited labeled data and structured or unstructured clinical notes.

Keywords:
free-text radiology reportlarge language modelsmetastases detectionnatural language processingsynthetic data generationtargeted data augmentation

More Related Videos

Using Computer-based Image Analysis to Improve Quantification of Lung Metastasis in the 4T1 Breast Cancer Model
08:32

Using Computer-based Image Analysis to Improve Quantification of Lung Metastasis in the 4T1 Breast Cancer Model

Published on: October 2, 2020

6.0K
Pathological Analysis of Lung Metastasis Following Lateral Tail-Vein Injection of Tumor Cells
08:54

Pathological Analysis of Lung Metastasis Following Lateral Tail-Vein Injection of Tumor Cells

Published on: May 20, 2020

8.8K

Related Experiment Videos

Last Updated: May 27, 2025

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
07:15

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model

Published on: August 16, 2020

6.6K
Using Computer-based Image Analysis to Improve Quantification of Lung Metastasis in the 4T1 Breast Cancer Model
08:32

Using Computer-based Image Analysis to Improve Quantification of Lung Metastasis in the 4T1 Breast Cancer Model

Published on: October 2, 2020

6.0K
Pathological Analysis of Lung Metastasis Following Lateral Tail-Vein Injection of Tumor Cells
08:54

Pathological Analysis of Lung Metastasis Following Lateral Tail-Vein Injection of Tumor Cells

Published on: May 20, 2020

8.8K

Area of Science:

  • Medical informatics
  • Natural Language Processing
  • Computational oncology

Background:

  • Accurate identification of metastatic sites in cancer patients is vital for treatment and diagnosis.
  • Automating metastasis detection from electronic health records (EHRs) is challenging due to data complexity.
  • Limited availability of expert-annotated data hinders the development of robust automated systems.

Purpose of the Study:

  • To automate the detection of metastatic sites from radiology reports using advanced natural language processing (NLP) techniques.
  • To leverage large language models (LLMs) for generating synthetic training data to overcome limitations of scarce labeled data.
  • To evaluate the efficacy of targeted data augmentation strategies and compare performance across different report structures.

Main Methods:

  • Prompting Llama3, an instruction-tuned LLM, to generate synthetic radiology report data for training.
  • Adapting BERT, a pretrained language model, for metastasis detection using expanded datasets.
  • Implementing and comparing three targeted data augmentation techniques against standard methods.
  • Analyzing metastasis identification accuracy using institutionally standardized versus non-structured reports, incorporating patient history with LoRA tuning.

Main Results:

  • Data augmentation using Llama3-generated synthetic data improved average F1-scores for lung, liver, and adrenal gland metastases by 2.3, 3.5, and 3.9 points, respectively.
  • Targeted data augmentation yielded comparable or superior performance to vanilla augmentation while being more computationally efficient.
  • Including patient history with a customized model architecture reduced the performance gap between standardized and non-structured reports from 7.3 to 4.5 F1-score points.

Conclusions:

  • LLMs can generate high-quality synthetic clinical data for automating metastasis detection, even without task-specific fine-tuning.
  • Targeted data augmentation and LLM-generated data offer a broadly applicable and computationally efficient solution for cancer progression analysis.
  • The developed approach enables large-scale, low-cost spatio-temporal extraction of cancer progression patterns, adaptable across different institutional data structures.