{"data":{"kind":"file","path":"README.md","version_id":"nmfk3arzh7kgz826ng9sp9jv","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2650,"modified_at":"2026-02-01T00:40:55.442000","content_hash":"c8449f22bdcfa39745c5b765f19ae7f66a67d1dd5eda0d0cbdb45b245e76ba61"},"entries":[],"content":"# OpenMed RadReport Environment\n\nRadiology Report Summarization environment for RL fine-tuning - generating clinical impressions from chest X-ray findings using MIMIC-CXR reports.\n\n## Task Description\n\nGiven the findings section of a radiology report, generate a concise clinical impression (summary/diagnosis). This is a critical task for:\n- Reducing radiologist workload\n- Ensuring consistent report quality\n- Training AI assistants for clinical documentation\n\n## Dataset\n\n- **Source**: [tgrex6/mimic-cxr-reports-summarization](https://huggingface.co/datasets/tgrex6/mimic-cxr-reports-summarization)\n- **Original**: MIMIC-CXR (Beth Israel Deaconess Medical Center)\n- **Train**: 91,544 reports\n- **Validation**: 2,000 reports\n- **Format**: Text-only (findings → impression)\n\n## Reward Structure\n\n| Component | Weight | Description |\n|-----------|--------|-------------|\n| Similarity | 50% | ROUGE-L-like score comparing to reference impression |\n| Clinical Quality | 35% | Appropriate medical terminology and structure |\n| Thinking | 15% | Quality of clinical reasoning in `<think>` tags |\n\n## Example\n\n**Input:**\n```\nClinical Background:\nShortness of breath\n\nFindings:\nThe lungs are clear without focal consolidation, pleural effusion, or\npneumothorax. The cardiomediastinal silhouette is within normal limits.\nNo acute osseous abnormality is identified.\n```\n\n**Expected Output:**\n```\n<think>\nThe findings describe normal lung fields with no consolidation, effusion, or\npneumothorax. The heart size is normal. No bone abnormalities. Given the\nclinical context of shortness of breath, the chest X-ray shows no acute\ncardiopulmonary cause for the symptoms.\n</think>\n\nNo acute cardiopulmonary abnormality.\n```\n\n## Usage\n\n```python\nfrom OpenMed_RadReport import load_environment\n\nenv = load_environment()\n```\n\n## Citation\n\n```bibtex\n@article{johnson2019mimic,\n  title={MIMIC-CXR, a de-identified publicly available database of chest radiographs with free-text reports},\n  author={Johnson, Alistair EW and Pollard, Tom J and Berkowitz, Seth J and Greenbaum, Nathaniel R and Lungren, Matthew P and Deng, Chih-ying and Mark, Roger G and Horng, Steven},\n  journal={Scientific Data},\n  volume={6},\n  number={1},\n  pages={317},\n  year={2019},\n  publisher={Nature Publishing Group}\n}\n\n@inproceedings{zhang2020radiology,\n  title={When Radiology Report Generation Meets Knowledge Graph},\n  author={Zhang, Yixiao and Wang, Xiaosong and Xu, Ziyue and Yu, Qihang and Yuille, Alan and Xu, Daguang},\n  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},\n  year={2020}\n}\n```\n\n## License\n\nMIT (dataset), PhysioNet License (original MIMIC-CXR)\n","encoding":"utf-8","truncated":false,"total_bytes":2650},"status":null}