Related Experiment Video
Updated: Jul 16, 2026
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
Large language model ensemble for automated TNM staging from radiology reports
Wen-Chao Yeh1,2, Yi-Shin Chen1, Wen-Lian Hsu1,3
1Institute of Information Systems and Applications, National Tsing Hua University, Hsinchu City, 300, Taiwan.
Bioinformatics (Oxford, England)
|July 14, 2026
Summary
Automated TNM staging for lung cancer using large language models (LLMs) achieved top ranks in a recent challenge. These AI systems improve accuracy and efficiency in cancer staging from radiology reports.
Area of Science:
- Artificial Intelligence
- Natural Language Processing
- Medical Informatics
Background:
- Accurate TNM staging of lung cancer is vital for patient treatment and prognosis.
- Manual staging is labor-intensive and prone to variability.
- Large language models (LLMs) present a promising avenue for automating this process.
Purpose of the Study:
- To develop and evaluate automated systems for TNM staging using LLMs.
- To compare the performance of different LLM-based approaches.
- To enhance the interpretability and clinical reasoning in automated cancer staging.
Main Methods:
- Developed two complementary systems for automated TNM staging from radiology reports.
- System I utilized GPT-4o with few-shot learning and multi-step voting.
- System II integrated GPT-4o and Gemini-2 via the DSPy framework with MIPROv2 optimization.
Main Results:
- Achieved first and second place in the NTCIR-18 RadNLP 2024 English main task.
- Demonstrated superior performance in T, N, and M classification with high accuracies.
- System I achieved joint accuracy of 0.6543, System II achieved 0.6296.
Conclusions:
- LLM-based systems can automate TNM staging effectively.
- The developed systems show competitive performance against established benchmarks.
- This work highlights the potential of AI in clinical decision support for oncology.
