DAMO's Abdominal CT AI Hits Science: One Model Detects 146 Diseases, Beats 23 of 26 Radiologists
量子位 · wechat · 2026-09-18
Alibaba DAMO Academy, with Zhejiang University First Hospital and other hospitals worldwide, published DAMO-RADAR in Science—described as the first expert-level generalist medical imaging AI, covering 18 abdominal anatomical structures and 146 diseases on CT. Validation: mean AUC 0.913 on 39K internal consecutive cases; 0.895 on 24K+ external cases from 8 hospitals; 0.904 on 27K emergency cases never included in training; and 0.891–0.984 on four cancers judged against pathology ground truth. It outperformed 23 of 26 radiologists from 14 hospitals; with AI assistance, doctors gained 10% sensitivity and cut reading time by over 30%. Method: no per-slice annotation—vision-language contrastive learning from CT-report pairs, with organ-level fine-grained alignment (decomposing each CT into organ units matched to report segments) and adaptive contrastive modeling informed by medical priors. Model, code and framework are fully open-sourced, extensible to MRI, PET and ultrasound.
More from Research
- PixelBench: testing multimodal models pixel-by-pixel instead of with QA — bdsqlsz · 2026-11-04
- François Chollet: modeling brain wiring as an engineering problem constrained by genome size and time — CatAstro_Piyush · 2026-11-04
- OffsecCyber dataset combines HackerOne reports and CVEs to train pentesting LLMs — AccBalanced · 2026-11-04
- Monofact rate in training data governs LLM scientific discovery, preprint finds — CatAstro_Piyush · 2026-11-04
- qwenTPU compiles a TPU into Qwen3.5 weights, achieving exact integer matmul — CatAstro_Piyush · 2026-11-04
- Jev model improves ICL demonstration retrieval via distillation into a lightweight MLP — kalyan_kpl · 2026-11-04