Open cancer model Merlin-Cancer-Super hits tumors in all 34 public CTs, best Dice 0.83
MaziyarPanahi · x · 2026-11-20
Developer @MaziyarPanahi ran the open-source Merlin-Cancer-Super model (Johns Hopkins + Stanford) on 34 public CT scans of women with uterine cancer, with no hints — and it touched every tumor:
- Best case lands inside the radiologist's outline, Dice 0.83
- It drew 30 of the 34 tumors smaller than the radiologist did
- The model is built on R-Super (MICCAI 2025 Best Paper), which uses an LLM to turn radiology reports into per-voxel supervision, covering detection and segmentation of tumors across 9 organs; the team claims it beats prior public models and leading VLMs
- Runs on a single GPU; weights are CC BY-NC-ND (non-commercial), scans from TCIA, and the author cautions public data may overlap pretraining — research only, not medical advice
A transparent third-party validation worth noting for open medical imaging AI.
Related event: Open-Source Cancer Model Detects Tumors in All 34 Test CTs(4 posts)→
More from Models
- 10 hands-on lessons on what actually happens inside an AI after you type a prompt — techartist_ · 2026-11-20
- Redwood questions Google reusing safety results from weaker model for stronger Carbon — YvesMulkers · 2026-11-20
- GTE-MLM sparse embedding models now natively supported in Transformers v5 — tomaarsen · 2026-11-20
- Reddit user mocks ChatGPT voice mode replying only 'Mm-hmm' at $20 for 3 hours — JackieFiest · 2026-11-20
- Xeophon discusses the ECI benchmark's creation and what works with its creators — xeophon · 2026-11-20
- GPT-6.2 Sol and Astra top RWE-Bench on cost-efficiency, moving the Pareto frontier — Jsevillamol · 2026-11-20