Project Imaging-X:面向基础模型开发的 1000+ 个开放医学影像数据集调查
计算机视觉与模式识别
2026-03-31 v1 人工智能
摘要
基础模型在各个领域和任务中展现出巨大成功,主要归因于大规模、多样且高质量数据集的蓬勃发展。然而,在医学影像领域,由于依赖临床专家知识且受严格的伦理和隐私约束,数据集的精心策划和组装具有较高挑战,导致大规模统一医学数据集稀缺,阻碍了强大医学基础模型的开发。本文present了目前为止规模最大的医学影像数据集调查,涵盖超过 1000 个开放数据集,系统性目录其模态、任务、解剖部位、注释、局限性及整合潜力。我们的分析揭示了该景观在规模上相对有限,跨狭窄任务范围碎片化,以及在器官和模态方面的分布不均,这进而限制了现有医学影像数据集用于开发通用且稳健医学基础模型的实用性。为将碎片化转化为规模化,我们提出了一种以元数据为驱动的融合范式 (MDFP),该范式整合具有相同模态或任务的公开数据集,从而将多个小数据孤岛转化为更大、更连贯的资源。基于 MDFP,我们发布了一个交互式发现门户,实现端到端、自动化的医学影像数据集集成,并将所有调查数据集编译成统一的结构化表格,清晰概述其关键特征并提供参考链接,为社区提供一个易于访问和全面的存储库。通过描绘当前图景并提供原则化的数据集整合路径,本次调查为扩展医学影像语料库、加快数据发现、更规范的数据集创建以及更强大的医学基础模型提供了实用路线图。
引用
@article{arxiv.2603.27460,
title = {Project Imaging-X: A Survey of 1000+ Open-Access Medical Imaging Datasets for Foundation Model Development},
author = {Zhongying Deng and Cheng Tang and Ziyan Huang and Jiashi Lin and Ying Chen and Junzhi Ning and Chenglong Ma and Jiyao Liu and Wei Li and Yinghao Zhu and Shujian Gao and Yanyan Huang and Sibo Ju and Yanzhou Su and Pengcheng Chen and Wenhao Tang and Tianbin Li and Haoyu Wang and Yuanfeng Ji and Hui Sun and Shaobo Min and Liang Peng and Feilong Tang and Haochen Xue and Rulin Zhou and Chaoyang Zhang and Wenjie Li and Shaohao Rui and Weijie Ma and Xingyue Zhao and Yibin Wang and Kun Yuan and Zhaohui Lu and Shujun Wang and Jinjie Wei and Lihao Liu and Dingkang Yang and Lin Wang and Yulong Li and Haolin Yang and Yiqing Shen and Lequan Yu and Xiaowei Hu and Yun Gu and Yicheng Wu and Benyou Wang and Minghui Zhang and Angelica I. Aviles-Rivero and Qi Gao and Hongming Shan and Xiaoyu Ren and Fang Yan and Hongyu Zhou and Haodong Duan and Maosong Cao and Shanshan Wang and Bin Fu and Xiaomeng Li and Zhi Hou and Chunfeng Song and Lei Bai and Yuan Cheng and Yuandong Pu and Xiang Li and Wenhai Wang and Hao Chen and Jiaxin Zhuang and Songyang Zhang and Huiguang He and Mengzhang Li and Bohan Zhuang and Zhian Bai and Rongshan Yu and Liansheng Wang and Yukun Zhou and Xiaosong Wang and Xin Guo and Guanbin Li and Xiangru Lin and Dakai Jin and Mianxin Liu and Wenlong Zhang and Qi Qin and Conghui He and Yuqiang Li and Ye Luo and Nanqing Dong and Jie Xu and Wenqi Shao and Bo Zhang and Qiujuan Yan and Yihao Liu and Jun Ma and Zhi Lu and Yuewen Cao and Zongwei Zhou and Jianming Liang and Shixiang Tang and Qi Duan and Dongzhan Zhou and Chen Jiang and Yuyin Zhou and Yanwu Xu and Jiancheng Yang and Shaoting Zhang and Xiaohong Liu and Siqi Luo and Yi Xin and Chaoyu Liu and Haochen Wen and Xin Chen and Alejandro Lozano and Min Woo Sun and Yuhui Zhang and Yue Yao and Xiaoxiao Sun and Serena Yeung-Levy and Xia Li and Jing Ke and Chunhui Zhang and Zongyuan Ge and Ming Hu and Jin Ye and Zhifeng Li and Yirong Chen and Yu Qiao and Junjun He},
journal= {arXiv preprint arXiv:2603.27460},
year = {2026}
}
备注
157 pages, 19 figures, 26 tables. Project repo: \url{https://github.com/uni-medical/Project-Imaging-X}