{"data":{"kind":"file","path":"README.md","version_id":"ewmamh9eyaofvc4re7z3nhj8","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":1467,"modified_at":"2026-08-14T01:20:11.432000","content_hash":"a44a16d9d429e1d41cd2cfbf87d01ff5c02f34dc7da27d97479700ac44d19737"},"entries":[],"content":"# X-Ray Diagnosis Environment\n\nA medical radiology environment that evaluates a model's ability to read and interpret X-ray images, identify pathologies, and provide structured clinical assessments.\n\n## Task Description\n\nThe model is presented with clinical vignettes describing X-ray findings across multiple imaging modalities (chest X-ray, skeletal, abdominal, dental). For each case, the model must:\n\n1. **Identify the imaging modality and anatomical region**\n2. **Describe key radiographic findings** systematically\n3. **Provide a differential diagnosis** (top 3 candidates ranked by likelihood)\n4. **Recommend appropriate follow-up** imaging or clinical actions\n\n## Scoring Rubric\n\n- **Finding Accuracy** (weight 0.30): Correct identification of primary pathology\n- **Differential Quality** (weight 0.25): Appropriateness and ranking of differential diagnoses\n- **Clinical Reasoning** (weight 0.25): Logical chain from findings to diagnosis\n- **Recommendation Appropriateness** (weight 0.20): Suitable follow-up actions\n\n## Difficulty Levels\n\n- **Level 0**: Classic textbook findings (pneumothorax, fracture)\n- **Level 1**: Common pathologies with subtle features\n- **Level 2**: Uncommon findings or multiple co-existing conditions\n- **Level 3**: Zebras - rare diagnoses requiring broad differential\n\n## Use Cases\n\n- RL training for medical reasoning\n- Evaluating clinical knowledge in LLMs\n- Benchmarking radiology AI assistants\n- Medical education assessment\n","encoding":"utf-8","truncated":false,"total_bytes":1467},"status":null}