One-for-All Adaptive Radiotherapy Planning Agent: A Foundation Framework for Daily CBCT-guided Radiotherapy

Area of Science:

• Medical Physics
• Artificial Intelligence in Medicine
• Radiotherapy

Background:

• Online adaptive radiotherapy (ART) is crucial for improving treatment accuracy by adjusting plans based on daily anatomical changes.
• Current ART workflows can be time-consuming, limiting widespread clinical adoption.
• Foundation models offer potential for unifying complex tasks in medical imaging and planning.

Purpose of the Study:

• To introduce a unified foundation-model-based agent for fully automated, online adaptive radiotherapy planning.
• To achieve complete, treatment-specific planning directly from daily cone-beam CT (CBCT) within a clinically relevant timeframe.
• To enable a human-in-the-loop framework for clinician oversight and intervention.

Main Methods:

• Development of the "One-for-All" Adaptive Radiotherapy Planning Agent, a foundation-model-based system.
• Autonomous prediction of planning components: synthetic CT generation, multimodal alignment, and tumor/organ segmentation from daily CBCT.
• Intelligent execution of clinical plan design leveraging predicted components.

Main Results:

• The agent completes online adaptive planning in under two minutes.
• Achieved clinically acceptable accuracy and plan quality comparable to reference plans across various cancer sites (head-and-neck, lung, abdominal, prostate) and modalities (photon, proton therapy).
• Target dose errors (D98) were generally within 2.0 Gy of reference plans.

Conclusions:

• The One-for-All agent demonstrates the potential of a unified foundation-model approach for automated online adaptive radiotherapy.
• The system offers a fast, scalable, and automated solution for diverse clinical scenarios.
• This technology facilitates efficient and accurate adaptive treatment planning with clinician involvement.