AI Plans Radiation Therapy in Two Minutes express
Each morning, a navigator plots a route, bypassing traffic jams and roadblocks. Similarly, before a radiation therapy session, a new AI agent recalculates the irradiation path. From a fresh CT scan, it finds the tumor and healthy organs on its own in two minutes, then calculates how to direct the radiation beam to hit the target and spare surrounding tissues.
The agent is built on a large neural network trained on numerous examples. It reconstructs the body's internal structure from the degree of beam attenuation (somewhat like spectroscopy) and precisely outlines the tumor boundaries. The plans match expert accuracy, with dose deviations staying within safe limits. The method was tested on head, lung, abdominal, and prostate cancers using both photon and proton therapy, and it proved reliable in all cases. Thus the legacy of radiotherapy, pioneered by Marie Curie and bolstered by the discoveries of William Lawrence Bragg, now gains a digital navigator that turns hours of calculation into minutes.
Original abstract
as written by the authors arXiv:2607.14870
In this work, we introduce the One-for-All Adaptive Radiotherapy Planning Agent, a unified foundation-model-based system that performs complete, treatment-specific online adaptive planning directly from daily cone-beam CT in under two minutes. The agent first autonomously predicts all essential planning components, including synthetic CT generation, multimodal alignment, and tumor/organ segmentation. It then intelligently leverages these outputs to execute the final clinical plan design, providing a comprehensive, automated solution for daily treatment. We also demonstrate that the agent enables clinicians to define planning with intent and intervene at critical decision points, ensuring a "human-in-the-loop" framework that generates acceptable plans before final approval. Evaluated on multiple datasets spanning head-and-neck, lung, abdominal, and prostate cancers with both photon and proton therapy, the proposed framework achieves clinically acceptable accuracy and plan quality comparable to clinically generated treatment plans, with target dose errors (D98) generally within 2.0 Gy of the reference plan. The strong performance of the One-for-All agent highlights the promise of a unified foundation-model approach and opens opportunities for fast, scalable, and fully automated online adaptive radiotherapy across diverse clinical scenarios.
Surprisingly, the system handles tumors in various locations equally well, from the head to the prostate, without any reconfiguration. It works with both conventional radiation and proton therapy, like a universal navigator that’s equally at home plotting a route through city streets or off-road.
Such systems are reminiscent of the GLaDOS assistant from the video game Portal, except instead of lab experiments, it's precise radiation control to save lives.