LAW & ORDER: Adaptive Spatial Weighting for Medical Diffusion and Segmentation

Medical image analysis depends on accurate segmentation and controllable synthesis, but both tasks face severe spatial imbalance: lesions occupy small regions against large backgrounds. We study adaptive spatial weighting as a task-level design principle and instantiate it in two adapters. LAW learns per-pixel loss weights for mask-conditioned diffusion by modulating a ratio prior with a feature-dependent delta map, with normalization, clamping, and Dice regularization for stability. ORDER improves lightweight segmentation by adding selective bidirectional skip attention with stage-wise confidence gating. On held-out diffusion test sets, LAW lowers FID from 158.13±0.15 to 108.43±0.71 on Polyps, from 144.13±0.31 to 89.51±0.96 on KiTS19, and from 139.22±0.38 to 112.58±0.68 on BRISC, while improving held-out mask-recovery Dice from 0.681±0.013 to 0.825±0.003 on Polyps. When the resulting images are added to nnUNet training, downstream Polyps mDice rises from 71.7±0.4 to 74.1±0.8. On the cleaned Polyps segmentation protocol, the reported ORDER configuration reaches 76.3±1.9 mDice and 67.2±2.0 mIoU at 42K parameters and 0.11 GFLOPs, versus 70.3±1.5 mDice and 59.9±1.7 mIoU for matched MK-UNet. On BRISC under the same training recipe, ORDER reaches 77.4±0.8 mDice and 68.1±0.7 mIoU. These results position adaptive spatial weighting as a practical design idea for both medical diffusion and efficient segmentation.