Clinical Large-Scale Integrative Multimodal Benchmark (CLIMB)
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/DDVD233/climb
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为CLIMB，它将多个公共的临床数据集整合为一个统一的基准，专为开发和评估多模态医疗人工智能系统而设计。该数据集覆盖了不同临床领域的多种模态。它不仅包含了多种模态的数据，保留了元数据和临床报告，并使用统一的标签空间进行训练，还提供了一个封闭选择的问答版本（CLIMB-QA）以供评估。该数据集规模宏大，包含约451万名患者的样本，总数据量达到19.01太字节。其任务旨在开发和评估多模态医疗人工智能系统。
This dataset, named CLIMB, integrates multiple public clinical datasets into a unified benchmark tailored for the development and evaluation of multimodal medical artificial intelligence systems. It encompasses diverse modalities across various clinical domains. In addition to containing multimodal data, it preserves metadata and clinical reports, utilizes a unified label space for model training, and additionally offers a closed-choice question answering variant (CLIMB-QA) for evaluation workflows. This large-scale dataset holds approximately 4.51 million patient samples, with a total data volume of 19.01 terabytes. Its core purpose is to facilitate the development and evaluation of multimodal medical AI systems.
