CheXanatomy generates chest X-ray anatomy segmentation via next-token prediction, accepted at MICCAI 2026
denisparra · x · 2026-09-29
Researchers present CheXanatomy, a MICCAI 2026 paper that injects explicit anatomical knowledge into a pretrained vision-language model for chest radiograph segmentation.
- Instead of task-specific decoder heads, the model learns to produce segmentation masks via autoregressive next-token prediction.
- Scalable supervision comes from synthesizing realistic chest X-rays from CT volumes and forward-projecting CT labels into anatomically consistent 2D masks.
- Evaluated against a U-Net baseline on synthetic and real radiographs, with ablations on model scale.
- Paper, code, and an HF demo are all publicly available.
More from Research
- Judea Pearl recommends Bareinboim's talk on causal AI in the age of LLMs — yudapearl · 2026-09-29
- The overlooked async RL curation pitfall: task runtime differences skew sampling — auto_grad_ · 2026-09-29
- RLDM 2027 heads to Paris, July 6-9, with DeepMind's David Abel as program chair — dabelcs · 2026-09-29
- SNaP one-step posterior sampling runs 30x faster than iterative samplers — prof_kamilov · 2026-09-29
- No matched data? Climate study's two-stage pattern cuts humid-heat bias by 44-57% — bravo_abad · 2026-09-29
- MIT Team's AI Agents Uncover Design Principles That Keep Atom-Thin Graphene From Catastrophic Failure — ProfBuehlerMIT · 2026-09-29