GLI-AL (BraTS-GLI Anatomy-Lesion)
收藏资源简介:
GLI-AL是一个基于BraTS 2023-GLI训练队列构建的多模态脑胶质瘤MRI标注资源,由西安交通大学研究团队开发,旨在解决原始数据集中未系统标注共存白质高信号的问题。该资源包含1251个四模态MRI案例的统一八类解剖-病变标签,总体积达15.1亿体素,其中新增病变体素220万,数据来源于BraTS挑战赛的受控访问数据。创建过程通过专家标注与自动工具融合,建立了包含394例净化子集和857例扩展子集的分层结构。该资源主要应用于医学图像联合分割研究,支持在统一标签空间中同时监督健康脑组织与病变结构,有效缓解标签噪声对模型训练的干扰,并促进脑肿瘤MRI分析的可靠性和可重复性评估。
GLI-AL is a multi-modal glioma MRI annotated resource constructed based on the BraTS 2023-GLI training cohort, developed by the research team from Xi'an Jiaotong University. It aims to address the issue that coexistent white matter hyperintensities were not systematically annotated in the original dataset. This resource contains unified eight-category anatomical-lesion labels for 1251 four-modal MRI cases, with a total volume of 1.51 billion voxels, including 2.2 million newly added lesion voxels. The data is sourced from the controlled-access data of the BraTS Challenge. The creation process fuses expert annotation with automatic tools, and establishes a hierarchical structure consisting of 394 purified subsets and 857 expanded subsets. This resource is mainly applied to medical image joint segmentation research, supporting simultaneous supervision of healthy brain tissue and lesion structures in a unified label space, effectively alleviating the interference of label noise on model training, and promoting the reliability and reproducibility evaluation of brain tumor MRI analysis.




