IMed-361M
收藏资源简介:
IMed-361M是由上海人工智能实验室和四川大学计算机科学学院联合创建的交互式医学图像分割基准数据集。该数据集整合了来自多个数据源的640万张医学图像及其对应的361亿个密集标注的掩码,涵盖14种成像模态和204个分割目标。数据集的创建过程包括图像的标准化、自动生成密集掩码以及严格的质量控制和粒度管理。IMed-361M旨在解决现有数据集在规模、多样性和密集标注方面的不足,支持医学图像分割模型的训练和评估,特别是在临床环境中实现高精度和可扩展性的分割任务。
IMed-361M is an interactive medical image segmentation benchmark dataset jointly created by the Shanghai AI Laboratory and the School of Computer Science, Sichuan University. This dataset integrates 6.4 million medical images from multiple data sources and their corresponding 36.1 billion densely annotated masks, covering 14 imaging modalities and 204 segmentation targets. The construction process of the dataset includes image standardization, automatic dense mask generation, as well as strict quality control and granularity management. IMed-361M aims to address the shortcomings of existing datasets in terms of scale, diversity and dense annotation, supporting the training and evaluation of medical image segmentation models, especially enabling high-precision and scalable segmentation tasks in clinical settings.
IMIS-Benchmark 数据集概述
数据集简介
- 名称: IMed-361M
- 类型: 多模态交互式医学图像分割数据集
- 规模: 包含超过361百万个掩码
- 图像数量: 640万张
- 掩码数量: 273.4百万个(每张图像平均56个掩码)
- 成像模态: 14种
- 分割目标: 204个
- 分辨率: 83%的图像分辨率在256×256到1024×1024之间
- 多样性: 涵盖六个解剖组,大部分掩码覆盖面积小于图像的2%
数据集来源
- 创建方式: 通过预处理来自多个来源的110个医学图像数据集生成
- 公开方式: 数据集将在HuggingFace上公开
数据集结构
sh dataset ├── BTCV │ ├─ image │ │ ├── xxx.png │ │ ├── .... │ │ ├── xxx.png │ ├── label │ │ ├── xxx.npz │ │ ├── .... │ │ ├── xxx.npz │ ├── imask │ │ ├── xxx.npy │ │ ├── .... │ │ ├── xxx.npy │ └── dataset.json
相关论文
- 论文标题: Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline
- 作者: Junlong Cheng, Bin Fu, Jin Ye, Guoan Wang, Tianbin Li, Haoyu Wang, Ruoyu Li, He Yao, Junren Chen, JingWen Li, Yanzhou Su, Min Zhu, Junjun He
- 年份: 2024
- arXiv链接: https://arxiv.org/abs/2411.12814

- 1Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline上海人工智能实验室 · 2024年



