Research map / Reinforcement learning / Visual and representation RL
Medical World Model
Yijun Yang, Zhao-Yang Wang, Qiuping Liu, Shuwen Sun, Kang Wang, Rama Chellappa, Zongwei Zhou, Alan Yuille, Lei Zhu, Yu-Dong Zhang, Jieneng Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add the Assistant Model (tumor segmentation) to the heuristic evaluation (ablation) | MeWM · TACE protocol exploration (in-house) | F1-score: 49.13 -> 52.38 | Improved |
| add the Assistant Model (ablation) | MeWM · TACE protocol exploration (public) | Precision: 75.00 (+2.4) without AM vs 72.62 with | Mixed |
| add Combo Contrastive Learning (ablation) | MeWM · TACE protocol exploration (in-house) | F1-score: 50.97 -> 52.38 | Improved |
| add Combo Contrastive Learning (ablation) | MeWM · TACE protocol exploration (public) | F1-score: 62.90 -> 64.08 | Improved |
| simulate post-treatment tumors rather than predicting from pre-treatment CT alone | multimodal GPT policies · TACE protocol exploration (in-house) | F1-score: 41.97 -> 52.38 | Improved |
| give physicians MeWM recommended protocol | Physician w/ Pre-CT · TACE protocol exploration (in-house) | F1-score: 48.81 -> 61.51 | Improved |
| give physicians MeWM recommended protocol | Physician w/ Pre-CT · TACE protocol exploration (public) | F1-score: 71.43 -> 80.00 | Improved |
| use MeWM generative model instead of prior tumor synthesis methods | visual Turing test · Visual Turing test, R1 | specificity (lower is better): 91.67 -> 79.17 | Improved |
| use deep survival heuristic function instead of Cox model | survival risk regression · HCC-TACE-Seg | MSE: 0.3550 -> 0.2142 | Improved |
| use deep model features instead of radiomics-based Cox | survival stratification · HCC-TACE-Seg | c-index: 0.472 -> 0.752 | Improved |
| MeWM versus physician with pre-treatment CT on public data | MeWM alone · TACE protocol exploration (public) | F1-score: 71.43 -> 64.08 | Hurt |
Closest papers by meaning
- CLARITY: Medical World Model for Guiding Treatment Decisions by Simulating Context-Aware Disease Trajectories in Latent Space (ECCV 2026 · similarity 0.90)
- Medical LLMs as Medical World Models: Unified Policy-Dynamics Learning with Test-Time Search (NeurIPS 2026 · similarity 0.90)
- Language Agents for Hypothesis-driven Clinical Decision Making with Reinforcement Learning (ICLR 2026 · similarity 0.85)
- PulseMind: A Multi-Modal Medical Model for Real-World Clinical Diagnosis (AAAI 2026 · similarity 0.85)
- Forecasting Treatment Responses Over Time Using Recurrent Marginal Structural Networks (NeurIPS 2018 · similarity 0.85)
- MTBBench: A Multimodal Sequential Clinical Decision-Making Benchmark in Oncology (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
