Human-like AI-based Auto-Field-in-Field Whole-Brain Radiotherapy Treatment Planning With Conversation Large Language
Adnan Jafar1, An Qin1, Gavin Atkins1
1Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, MD, USA.
Arxiv
|January 8, 2026
Summary
This study introduces an automated radiotherapy planning pipeline using deep learning for parameter prediction and a large-language model for interactive refinement, significantly reducing planning time and maintaining clinical acceptability.
Area of Science:
- Medical Physics
- Artificial Intelligence in Radiation Oncology
Background:
- Whole-brain radiotherapy (WBRT) is a standard treatment, but its planning requires manual optimization.
- Automated Field-in-Field (Auto-FiF) tools in treatment planning systems still necessitate patient-specific adjustments and feedback-based refinement.
Purpose of the Study:
- To develop an automated WBRT planning pipeline integrating deep learning (DL) and large-language models (LLMs).
- To enable patient-specific hyperparameter prediction and interactive plan refinement for improved efficiency and quality.
Main Methods:
- A DL Hyperparameter Prediction model was trained on 55 WBRT cases using CTV and OAR geometric features.
- An LLM-based conversational interface (Whisper, GPT-4o) was used for interactive plan refinement based on voice feedback.
- Plan quality was assessed in 15 independent cases using clinical metrics and expert review.
Main Results:
- Fourteen of 15 DL-generated plans were clinically acceptable, with no significant differences in OAR or CTV dose metrics compared to manual plans.
- The automated pipeline achieved total workflow execution in approximately 7 minutes, a substantial reduction from manual planning (15 minutes).
- The conversational module effectively improved dose conformity and reduced hotspots in suboptimal plans.
Conclusions:
- The integrated DL and LLM pipeline offers an efficient and effective automated solution for WBRT planning.
- This approach maintains or improves plan quality while significantly reducing planning time and clinician workload.
- The system demonstrates potential for wider adoption in radiotherapy planning workflows.
