agentN0/usmle-step1-form31

该数据集名为USMLE Step 1 — Form 31, Form 30 & NBME Form 27，是一个医学领域的多项选择题库，专注于美国医师执照考试（USMLE）Step 1的练习表格。数据集包含从USMLE Step 1 Form 31、NBME Form 27和NBME Form 30中提取的结构化问题，总计581个问题（Form 31有183个问题，Form 27有198个问题，Form 30有200个问题），并附有121张经过SOTA裁剪的图像。每个问题记录包括唯一标识符、问题编号、所属表格ID、自动推断的医学主题、问题文本、选项数组（字母和文本）、正确答案字母、解释以及图像数组。图像经过多阶段计算机视觉处理提取，包括集合检测、布局感知评分、OCR验证和边界细化，以确保仅包含临床视觉内容（如X光片、CT、MRI、照片、组织学图像、图表等），没有全页截图或文本元素。数据集适用于医学教育、NLP研究和AI辅助学习场景。The dataset named USMLE Step 1 — Form 31, Form 30 & NBME Form 27 is a medical multiple-choice question bank focused on practice forms for the United States Medical Licensing Examination (USMLE) Step 1. It contains structured questions extracted from USMLE Step 1 Form 31, NBME Form 27, and NBME Form 30, totaling 581 questions (183 from Form 31, 198 from Form 27, and 200 from Form 30), accompanied by 121 SOTA-cropped images. Each question record includes a unique identifier, question number, form ID, auto-inferred medical subject, question text, an array of options (with letters and text), the correct answer letter, an explanation, and an array of images. The images are extracted using a multi-stage computer vision pipeline involving ensemble detection, layout-aware scoring, OCR validation, and boundary refinement to isolate only clinical visual content (such as X-rays, CT scans, MRIs, photos, histology images, diagrams, etc.), without full-page screenshots or text elements. The dataset is suitable for medical education, NLP research, and AI-assisted learning scenarios.