Latent diffusion predicts 3d radiation doses for faster treatment planning
Learning the Dose Manifold: A Latent Diffusion Framework for Volumetric Dose Prediction in Automated Radiotherapy Treatment Planning
Computer Vision and Pattern Recognition
Summary
Planning radiation therapy for cancer patients is complex and varies a lot between doctors. The authors developed a new computer method called LaDose that predicts the full 3D radiation dose needed based on patient anatomy. This method compresses the data to handle 3D information efficiently on a single graphics card and can adjust treatment priorities without retraining. Tested on hundreds of real patient plans, it produces accurate dose predictions quickly and was rated good enough for clinical use by experienced doctors.
What this means in practice
- •For radiation therapy planners: Generate accurate 3D radiation dose predictions for automated treatment plan creation, reducing variability and speeding up plan generation.
- •For medical imaging software developers: Integrate a latent diffusion-based 3D dose prediction model into radiotherapy planning software for improved clinical workflow and dose modulation.$Commercial implications: Enables enhanced treatment planning software products for hospitals by improving dose prediction accuracy and computation speed.
Authors
J. Boyer-Chammard, R. Vauclin, M. C. Biston, P. Maury, M. Gau, S. Stathakis, V. Grégoire, N. Paragios, E. Deutsch, N. Komodakis
Abstract
Automated treatment planning (ATP) aims to reduce the unwarranted inter-planner variability of manual planning, while improving treatment quality and efficiency. Most ATP pipelines rely on the prediction of the achievable three-dimensional dose distribution for a given anatomy and the subsequent optimization of machine parameters to reproduce it. Diffusion-based methods are largely slice-wise due to the high dimensionality of radiotherapy data, and assume a one-to-one mapping between anatomy and optimal dose despite the multi-objective nature of treatment planning. They are evaluated on their ability to match reference doses and rarely on their integration in ATP frameworks. We present LaDose, a latent diffusion framework for fully volumetric dose prediction. An autoencoder learned on dose distributions, trained so that perturbed latents decode to plausible doses, compresses dose maps 128-fold and enables 3D diffusion on a single GPU. Prescription values are represented as exact scalar vectors that bypass the image encoder, avoiding resampling and latent compression. Multiscale classifier-free guidance allows for the modulation of clinical trade-offs at inference, without retraining or additional labeled data. On 900 clinically delivered head-and-neck volumetric modulated arc therapy plans from five centers, LaDose reached a median absolute error of 1.18 Gy (1.4--1.8\% of the prescription) and was the only one of six frameworks free of systematic bias with respect to the reference distribution. Guidance scales acted predictably on target coverage and organ-at-risk sparing. Integrated with direct aperture optimization, the pipeline generated plans in 15.2 minutes; in a blinded review, senior clinicians rated 14 of 15 plans acceptable without editing.