Related Experiment Video
Updated: Jun 9, 2025
08:17
Dynamic Lung Tumor Tracking for Stereotactic Ablative Body Radiation Therapy
Published on: June 7, 2015
15.7K
LLM-driven multimodal target volume contouring in radiation oncology
Yujin Oh1, Sangjoon Park2,3, Hwa Kyung Byun4
1Department of Radiology, Massachusetts General Hospital (MGH) and Harvard Medical School, Boston, MA, USA.
Nature Communications
|October 25, 2024
Summary
A new AI model, LLMSeg, uses large language models (LLMs) to improve radiation therapy target volume delineation by integrating image and text data. This multimodal approach enhances accuracy and efficiency in cancer treatment planning.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Radiation Oncology
Background:
- Target volume contouring in radiation therapy is complex, requiring integration of both imaging and clinical text data.
- Current AI models often struggle with multimodal data integration for precise target delineation.
Purpose of the Study:
- To introduce LLMSeg, an LLM-driven multimodal AI for 3D context-aware target volume delineation in radiation oncology.
- To assess LLMSeg's performance in challenging scenarios like breast cancer radiotherapy, external validation, and data-insufficient environments.
Main Methods:
- Developed LLMSeg, a multimodal AI leveraging large language models (LLMs) to fuse image and text-based clinical information.
- Validated LLMSeg in breast cancer radiotherapy, focusing on external validation and data-scarce conditions.
Main Results:
- LLMSeg demonstrated significantly improved performance over conventional unimodal AI models.
- The multimodal AI exhibited robust generalization capabilities and data efficiency in validation studies.
Conclusions:
- LLM-driven multimodal AI offers a promising approach for enhancing target volume delineation in radiation oncology.
- LLMSeg shows potential for real-world clinical application, particularly in data-limited settings.
