Synthetic liver SBRT · LIVER_50

**This CT was generated by a model. No patient was scanned, and there is no real case behind it.** Produced with NVIDIA MAISI (3D latent diffusion) via MONAI, conditioned on a segmentation from the model's own candidate set, at 512×512×512 and 1 mm isotropic — the combination MAISI's own quality check passes. The Hounsfield units are the model's and have not been replaced; bone above 700 HU was stretched back into a cortical range to undo the training clip at 1000. Treat the HU as plausible rather than measured. There is no scanner behind this image, so no CT-to-density calibration of yours applies to it, and because the volume has no real counterpart the error in its densities cannot be quantified at all. Every plan submitted here is scored against the same geometry, so the ranking compares planners — but nothing learned from this case is a statement about a real patient. Target 6.6 cc in a 1438 cc liver. Liver_GTV is supplied rather than derived by the planner, so every plan is scored on the same denominator. Prescription is 50 Gy in 5 fractions per NRG/RTOG 1112; that trial lowered the dose when the liver mean could not be met, and this challenge does not. Target check: 6.6 cc, 1 piece(s) generated, +18.5 HU against the liver around it — below the platform's threshold of 20 HU contrast; the lesion is hard or impossible to see on the image.