Sign in
Not yet recruitingNo phase

Multimodal LLM for Breast Cancer Radiotherapy Target Auto-Contouring and Planning

Key facts

Who runs it
Xiaohui Zhao
Where
China1 site in 1 countryChoose a country to see its sites
Trial code
Status
Not yet recruiting
People taking part
51,000 people, planned
Started
9 Nov 2026
Expected to end
31 Dec 2027
Type of study
Observational, 3 groups
Treatments tested
  • Multimodal LLM-assisted target volume auto-segmentation and intelligent radiotherapy planning system
Record updated
17 November 2026

Who can take part

Sex
Women and men
Healthy volunteers
Cannot take part

The registry criteria, word for word

Show all 8 criteriaShow less

Who can take part

  • Patients with pathologically confirmed breast malignant tumors who require postoperative adjuvant radiotherapy following radical surgery
  • Availability of complete pathology reports
  • Availability of complete imaging examinations and reports (such as CT, MRI, ultrasound, X-ray)
  • Availability of standardized clinical medical records

Who cannot take part

  • Patients receiving salvage or palliative radiotherapy
  • Patients receiving accelerated partial breast irradiation (APBI)
  • Patients with primary multiple cancers
  • Patients with missing data

Where is this trial?

It has sites in one country. Choose a country to see its sites.

About the trial

What the trial wants to find out

Breast cancer remains the most prevalent malignancy among women worldwide, with a continuously increasing incidence. As a crucial component of comprehensive cancer treatment, radiotherapy plays a key role in reducing local recurrence rates and improving overall survival outcomes. However, traditional radiotherapy target volume contouring and treatment planning design rely heavily on clinician expertise, presenting challenges such as prolonged processing time, significant inter-observer variability, and high subjectivity in dose optimization. Although artificial intelligence has made substantial progress in automated contouring and planning, conventional single-modal models still face major limitations. These models struggle to effectively integrate clinical text, tumor biological characteristics, and dose distribution factors, resulting in restricted generalizability. Therefore, developing an intelligent radiotherapy assistance system capable of multimodal data fusion is of high clinical significance to enhance precision and efficiency in breast cancer radiotherapy.

Read full original descriptionShow less

This study aims to develop an innovative breast cancer radiotherapy assistance system based on a multimodal large language model. By integrating multimodal imaging data (including CT, MRI, and ultrasound), pathology reports, and structured electronic health record text, the system realizes end-to-end intelligence from target volume contouring to treatment plan generation. Employing a multicenter, ambidirectional cohort design, the study collects comprehensive clinical data from breast cancer patients who have undergone breast-conserving surgery or mastectomy. Technologically, the system innovatively combines the natural language processing capabilities of large language models with the image analysis strengths of deep learning, utilizing attention mechanisms to achieve deep fusion and interactive comprehension of heterogeneous medical data. Notably, the system breaks through traditional radiotherapy planning workflows by jointly optimizing target contouring and dose planning as a unified pipeline, applying reinforcement learning algorithms to dynamically balance target volume coverage with organ-at-risk protection. The outcomes of this study will provide an important paradigm for the clinical application of AI in radiation oncology and advance the field of precision radiotherapy.

The 3 groups in the trial

  • Retrospective Training and Internal Validation Cohort

    Retrospective collection of clinical data, multimodal images (MRI, CT, ultrasound, X-ray), pathology reports, and radiotherapy plans from patients with breast cancer treated at Sun Yat-sen Memorial Hospital between January 1, 2002, and October 31, 2025. This cohort (n=35,000) is used for model training and internal validation of the multimodal LLM-assisted automated contouring and radiotherapy planning system.

    Treatment: Multimodal LLM-assisted target volume auto-segmentation and intelligent radiotherapy planning system

  • Retrospective External Validation Cohort

    Retrospective collection of data from external collaborative centers (Sun Yat-sen Memorial Hospital Shenzhen Hospital, Zhoukou Central Hospital, and Tongji Hospital, Tongji Medical College, Huazhong University of Science and Technology) between January 1, 2002, and October 31, 2025 (total n=15,000). This cohort is used to evaluate the external generalizability and robustness of the multimodal AI model.

    Treatment: Multimodal LLM-assisted target volume auto-segmentation and intelligent radiotherapy planning system

  • Prospective Test Cohort

    Prospective collection of data from patients with breast cancer treated at Sun Yat-sen Memorial Hospital starting from January 1, 2026 (n=1,000). This cohort undergoes double-blind evaluation to prospectively test the performance of the multimodal LLM-driven target volume auto-segmentation and intelligent radiotherapy planning system in clinical practice.

    Treatment: Multimodal LLM-assisted target volume auto-segmentation and intelligent radiotherapy planning system

Questions to ask the study team

  1. How often do I need to come in, and for how long each time?
  2. Are travel costs paid?
  3. Can I keep my own doctor?
  4. What happens if I want to stop?

Contact the study team: see the registry record → Contacts and Locations

Print a page for my doctor

Data and sources

Registry record
Listed on ClinicalTrials.gov.How to verify a trial in the registry
Official title
Multimodal Data Fusion-Based Artificial Intelligence Large Language Model for Automated Target Volume Contouring and Intelligent Treatment Planning in Breast Cancer Radiotherapy: A Multicenter Ambidirectional Cohort Study
Record updated
17 November 2026