JHU & Stanford release Merlin-Cancer-Super, beating SOTA and VLMs on 9-organ tumor segmentation
MaziyarPanahi · x · 2026-11-04
Stanford and Johns Hopkins teams released Merlin-Cancer-Super on Hugging Face, built on R-Super (MICCAI 2025 Best Paper Award, top 2). It surpasses state-of-the-art public AI models and leading VLMs at detecting and segmenting tumors across 9 organs: bladder, gallbladder, spleen, esophagus, stomach, duodenum, prostate, uterus, adrenal glands.
Method highlights:
- R-Super turns radiology reports into per-voxel supervision: an LLM extracts tumor info from reports first
- New Volume Loss & Ball Loss losses push segmentations to match reports in tumor count, diameters, and locations
Trained on Merlin Plus (Stanford, 25,494 CT scans with longitudinal metadata and per-voxel annotations), openly available on HF. Research on public data, not medical advice.
More from Multimodal
- PianoBench: open audio models top out at 52.8% on musical reasoning — kastnerkyle · 2026-11-05
- Runway Dev lands on Claude Code to auto-integrate AI video, image and audio APIs — runwayml · 2026-11-05
- Dynamic VRAM makes MiniMax H3 VAE loads take minutes, reports Reddit user — poursoul · 2026-11-04
- Tencent launches Hy4: first Hunyuan model with native vision and computer use — Aiden_Tech_Ai · 2026-11-04
- Identity-aware 4K image editing: a MiniMax H3 single-frame edit workflow — arthan1011 · 2026-11-04
- Meituan launches LongCat-2.5: 1.6T-param native multimodal agentic model — thisdudelikesAI · 2026-11-04