Research map / Vision-language and multimodal / Cross-modal learning
PulseMind: A Multi-Modal Medical Model for Real-World Clinical Diagnosis
Jiao Xu, Junwei Liu, Jiangwei Lao, Qi Zhu, Yunpeng Zhao, Congyun Jin, Shinan Liu, Zhihong Lu, Lihe Zhang, Xin Chen, Jian Wang, Ping Wang
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 MediScope multi-turn multimodal dataset to SFT data (ablation) | PulseMind-72B · PulseMind Benchmark | avg win rate (%): 26.4 -> 65.2 | Improved |
| add MediScope to SFT data (ablation) | PulseMind-72B · MedXpertQA-MM | accuracy: 34.9 -> 36.5 | Improved |
| add MediScope to SFT data (ablation) | PulseMind-72B · MMMU Health & Medicine | accuracy: 67.3 -> 68.1 | Improved |
| add RL stage (CRPO) after SFT (ablation) | PulseMind-72B · PulseMind Benchmark | avg win rate (%): 65.2 -> 76.0 | Improved |
| add RL stage after SFT (ablation) | PulseMind-72B · MedXpertQA-MM | accuracy: 36.5 -> 36.6 | No clear effect |
| add RL stage after SFT (ablation) | PulseMind-72B · MMMU Health & Medicine | accuracy: 68.1 -> 69.4 | Improved |
| use CRPO comparison-based reward instead of GRPO absolute-score reward (ablation) | PulseMind-72B · PulseMind Benchmark | avg win rate (%): 54.7 -> 76.0 | Improved |
| use CRPO instead of GRPO (ablation) | PulseMind-72B · MMMU Health & Medicine | accuracy: 66.7 -> 69.4 | Improved |
| use CRPO instead of GRPO (ablation) | PulseMind-72B · SLAKE | accuracy: 85.2 -> 85.6 | No clear effect |
| use relative (pairwise) scoring instead of absolute scores for GPT judge (ablation) | GPT-based evaluator · PulseMind Benchmark | consistency with experts (%): 51.5% -> 86.1% | Improved |
| use absolute 5-point scoring to rank models | GPT-based evaluator · PulseMind Benchmark | score spread: 4.01 to 4.35 average | Hurt |
| use PulseMind-72B instead of Qwen2.5VL-72B | PulseMind-72B · VQA-RAD | accuracy: 80.3 -> 87.1 | Improved |
Closest papers by meaning
- Multi-modal Medical Diagnosis via Large-small Model Collaboration (CVPR 2025 · similarity 0.89)
- QoQ-Med: Building Multimodal Clinical Foundation Models with Domain-Aware GRPO Training (NeurIPS 2025 · similarity 0.88)
- MedAgent-Pro: Towards Evidence-based Multi-modal Medical Diagnosis via Reasoning Agentic Workflow (ICLR 2026 · similarity 0.88)
- MedTVT-R1: A Multimodal LLM Empowering Medical Reasoning and Diagnosis (CVPR 2026 · similarity 0.88)
- MedAtlas: Evaluating LLMs for Multi-Round, Multi-Task Medical Reasoning Across Diverse Imaging Modalities and Clinical Text (AAAI 2026 · similarity 0.87)
- RAD: Towards Trustworthy Retrieval-Augmented Multi-modal Clinical Diagnosis (NeurIPS 2025 · similarity 0.87)
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.
