The Flow has not summarised this story yet — read it at arXiv Computer Vision.
BEAM3R is a dose estimation framework that operates in beam’s-eye-view (BEV) for both photon and proton beamlet calculations. It uses a Mamba‑3 state‑space depth‑sequence core combined with physics‑based transport conditioning, sharing a 2D CNN encoder‑decoder architecture to process per‑plane BEV slices. The system achieves high accuracy on the DoseRAD2026 challenge, with local gamma pass rates above 96% for CT‑based models and runtimes under 25 s for photon and 18 s for proton predictions.
A diffusion‑transformer framework was developed to denoise low‑statistics Monte Carlo dose distributions for intensity‑modulated proton therapy. Trained on 80 head‑and‑neck patient datasets, the model was evaluated on 10 additional head‑and‑neck cases and 30 patients with lung, breast, and prostate cancers, achieving mean absolute errors of 0.08–0.20 Gy and 3 %/2 mm gamma passing rates above 95 %. The denoised dose maps closely matched high‑statistics reference distributions in dose‑volume histogram metrics for clinical targets and organs at risk.
Dose-PlanNet is a physics-guided 3D deep learning architecture that predicts radiotherapy dose distributions for prostate cancer. In a prospective trial, the model achieved target coverage comparable to clinical plans while slightly reducing target homogeneity, yet it significantly improved high‑dose organ‑at‑risk sparing. Automated plans met clinical acceptance criteria in 11 of 14 Moderate Hypofraction and 9 of 12 SBRT plans, demonstrating that physics-informed deep learning can accelerate radiotherapy workflows without compromising dosimetric quality.
arXiv:2606. 29592v1 Announce Type: new Abstract: A central premise of autonomous scientific imaging is that smarter navigation, whether Bayesian, RL-based, or otherwise adaptive, is the principal lever for sample-efficient acquisition.
Dose-PlanNet is a physics-guided 3D deep learning model that predicts radiotherapy dose distributions for prostate cancer, aiming to automate complex treatment planning. In a prospective trial, the model achieved comparable target coverage while slightly reducing target homogeneity, yet it significantly improved high-dose organ‑at‑risk sparing. Automated plans met clinical acceptance criteria in most cases across both moderate hypofraction and stereotactic body radiation therapy arms.
A central premise of autonomous scientific imaging is that smarter navigation, whether Bayesian, RL-based, or otherwise adaptive, is the principal lever for sample-efficient acquisition. We present evidence to the contrary in scanning transmission electron microscopy (STEM), an atomic-resolution imaging modality whose every measurement deposits damaging electron dose.