Related Experiment Video
Updated: Jun 24, 2026
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
Leveraging Large Language Models to Integrate Clinical Knowledge and Machine Learning Predictions for Lymph Node
Hongying Yu1, Bing Liu2, Xian Zeng1
1Jiangsu Key Laboratory of Intelligent Medical Image Computing, School of Artificial Intelligence, Nanjing University of Information Science and Technology, Linjiang Building, No.219, Ningliu Road, Nanjing, 210044, China, 1 13291879390.
This study integrates large language models (LLMs) and machine learning (ML) to improve lung cancer lymph node metastasis (LNM) prediction. The novel framework enhances diagnostic accuracy by combining clinical knowledge from LLMs with data-driven ML patterns.
Area of Science:
- Oncology
- Medical Informatics
- Artificial Intelligence
Background:
- Lymph node metastasis (LNM) is crucial for lung cancer treatment decisions.
- Accurate preoperative LNM diagnosis remains a significant clinical challenge.
- Existing data-driven models often neglect valuable clinical knowledge.
Purpose of the Study:
- To enhance lung cancer lymph node metastasis (LNM) prediction accuracy.
- To investigate the integration of LLM-derived knowledge with data-driven patterns.
- To develop a novel framework combining LLMs and machine learning (ML) for LNM prediction.
Main Methods:
- An ensemble framework combining LLMs and ML models was developed.
- Three ML models were trained on clinical data; their outputs were fed into LLM prompts.
- Five LLMs (GPT-5.4, GPT-5.4-nano, DeepSeek-V3.2) were used with ensemble strategies for prediction.
Main Results:
- The proposed framework significantly outperformed base ML models.
- Achieved an area under the curve (AUC) of 0.781 and average precision of 0.420.
- Reasoning settings (English/Chinese) showed varied performance compared to non-reasoning English.
Conclusions:
- A novel knowledge-augmented strategy was presented for LNM prediction in lung cancer.
- Integration of LLM clinical knowledge with ML statistical patterns improves prediction.
- This offers a new paradigm for combining medical knowledge and patient data in clinical predictions.
