mugezhang/medical-diff-vqa-longitudinal

这是一个增强版的医学视觉问答数据集，专注于放射学领域的胸部X光片分析。数据集包含预计算的时间序列图像上下文信息，采用最近N=4策略选择上下文图像。数据经过严格筛选，仅保留差异类型的问题和具有时间序列上下文的病例。数据集包含123,980个问题，涉及19,410名独特患者，分为训练集(99,194)、验证集(12,349)和测试集(12,437)。每个问题包含问题文本、答案、2张参考图像和最多4张上下文图像(平均2.86张)。图像对象包含研究ID、DICOM ID、视图位置、研究日期和图像路径等元数据。数据集需要PhysioNet认证才能访问，主要用于研究目的。

This is an enhanced, filtered subset of Medical-Diff-VQA with pre-computed longitudinal imaging context. The dataset focuses on radiology chest X-ray analysis and employs a most recent N=4 strategy for context image selection. It contains only the difference subtype questions from patients with ≥1 additional study before the reference pair. The dataset consists of 123,980 questions from 19,410 unique patients, split into train (99,194), validation (12,349), and test (12,437) sets. Each question includes the question text, answer, 2 reference images, and up to 4 context images (average 2.86). Image objects contain metadata like study ID, DICOM ID, view position, study date, and image path. Access requires PhysioNet credentialing and is intended for research use only.