{"data":{"kind":"file","path":"README.md","version_id":"m72e90twjci9xliv477v5qqq","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":2736,"modified_at":"2026-01-14T16:53:24.692000","content_hash":"047ca572095bfbe18e876ba2bd7bbdebe65de114fe4bf6706618241e6960f31c"},"entries":[],"content":"# OpenMed RadQA\n\nRadiology-focused medical question answering environment for reinforcement learning fine-tuning.\n\n## Overview\n\nOpenMed RadQA is a specialized medical QA environment focused on radiology, medical imaging, and diagnostic procedures. It filters MedQA-USMLE questions to create a curated dataset of radiology-related clinical scenarios.\n\n## Task\n\nGiven a clinical scenario involving radiology or medical imaging with 4 multiple-choice options (A, B, C, D), select the most appropriate answer.\n\n**Format:**\n- Input: Clinical question with imaging findings and 4 options\n- Output: Single letter answer (A-D) wrapped in `\\boxed{}`\n- Reasoning: Enclosed in `<think>` tags\n\n## Dataset\n\n**Source:** MedQA-USMLE-4-options (GBaker/MedQA-USMLE-4-options)\n\n**Filtering:** Questions are filtered using radiology-specific keywords including:\n- Imaging modalities: X-ray, CT, MRI, ultrasound, PET, mammography\n- Radiology terms: radiograph, contrast study, interventional radiology\n- Clinical imaging: chest film, bone scan, angiography, doppler\n\n**Size:**\n- Training: ~800-1000 radiology questions (filtered from 10,200 total)\n- Evaluation: ~150-200 radiology questions (filtered from test split)\n\n## Clinical Importance\n\n- **Diagnostic Imaging:** Critical for accurate interpretation of radiological findings\n- **Patient Safety:** Appropriate imaging modality selection and contrast use\n- **Clinical Decision-Making:** Integration of imaging findings with clinical presentation\n- **Board Exam Preparation:** USMLE-style radiology questions\n\n## Example\n\n**Question:**\n```\nA 45-year-old man presents with sudden onset severe headache.\nNon-contrast CT scan of the head shows hyperdense material in the\nsubarachnoid space. What is the most likely diagnosis?\n\nA. Ischemic stroke\nB. Subarachnoid hemorrhage\nC. Subdural hematoma\nD. Epidural hematoma\n```\n\n**Expected Output:**\n```\n<think>\nThe patient presents with sudden severe headache (\"thunderclap\" headache).\nNon-contrast CT shows hyperdense material in subarachnoid space - this is\nclassic for acute blood in the subarachnoid space.\nA. Ischemic stroke would show hypodense area, not hyperdense\nB. Subarachnoid hemorrhage matches the imaging and presentation\nC. Subdural hematoma would be between dura and arachnoid, different location\nD. Epidural hematoma would be between skull and dura, lens-shaped\nThe answer is B.\n</think>\n\\boxed{B}\n```\n\n## Reward Structure\n\n- **Accuracy (80%):** Correct answer selection\n- **Thinking (15%):** Comprehensive clinical reasoning (300+ chars for full credit)\n- **Format (5%):** Proper `\\boxed{}` usage\n\n## Installation\n\n```bash\nprime env install maziyar/OpenMed_RadQA\n```\n\n## Version\n\n**v0.1.0** - Initial release with MedQA-USMLE radiology filtering\n","encoding":"utf-8","truncated":false,"total_bytes":2736},"status":null}