spatialmed/croissant_spatialmed

SpatialMed是一个容积医学多选问答基准测试数据集，包含8,864个经过筛选的多选题/是非题，覆盖9个源解剖数据集（如BraTS、KiTS、AMOS等），并分为4个空间推理任务（方向、范围、体积、比较）。每个问题都基于3-D NIfTI容积和三个正交的2-D RGB视图（轴向、矢状、冠状）。数据集提供了详细的文件夹布局、任务类型、模态信息、快速开始指南以及许可证信息。

SpatialMed is a volumetric medical multiple-choice question (MCQ) benchmark dataset comprising 8,864 filtered MCQ/yes-no questions across 9 source anatomical datasets (e.g., BraTS, KiTS, AMOS) grouped into 4 spatial-reasoning tasks (DIR, EXT, VOL, COMP). Each question is grounded in both a 3-D NIfTI volume and three orthogonal 2-D RGB views (axial, sagittal, coronal). The dataset includes detailed folder layout, task types, modalities, quick-start guide, and licensing information.