Siemens Healthineers' diffusion-LSTM model speeds up VMAT radiotherapy planning
One-shot fluence map generation cuts repeated re-optimization time for prostate cancer patients.
A diffusion-driven Learning-to-Optimize method is presented for end-to-end Volumetric Modulated Arc Therapy (VMAT) planning. A distribution-matching distilled diffusion model learns a clinically feasible manifold of fluence maps for one-shot generation, and an LSTM-based module learns gradient update dynamics to refine fluence maps toward prescribed dose objectives during inference. Experimental results on clinical and public prostate cancer cohorts show improved planning efficiency, flexibility, and machine deliverability over currently available end-to-end VMAT planners. The paper received an Early Accept at MICCAI 2026.
- A distilled diffusion model generates clinically feasible VMAT fluence maps in a single forward pass, replacing iterative re-optimization.
- An LSTM-based Learning-to-Optimize module refines fluence maps toward prescribed dose constraints during inference.
- Tested on clinical and public prostate cancer cohorts; early accepted at MICCAI 2026.
Why It Matters
Could slash radiotherapy planning from hours to minutes, freeing clinicians for more patients and reducing treatment delays.