Angelou0516/UCSF-PDGM
收藏资源简介:
UCSF-PDGM(加州大学旧金山分校术前弥漫性胶质瘤MRI数据集)的BraTS格式结构性子集,包含495名经组织病理学确诊的弥漫性胶质瘤患者的术前3T脑MRI数据。每位患者提供四种结构性MRI模态(T1、T1c、T2、FLAIR)和专家校正的肿瘤分割掩码,所有数据已配准到240×240×155的BraTS网格(1毫米各向同性,去颅骨)。数据集还包括临床元数据,如性别、年龄、WHO CNS分级、WHO-2021诊断、IDH状态、MGMT状态、总生存期等。肿瘤分割遵循BraTS标签约定:0为背景,1为非增强肿瘤核心/坏死,2为瘤周水肿,4为增强肿瘤。注意:约60%病例与BraTS 2021/2023重叠,需在独立基准测试中排除重叠部分以避免数据泄露。数据预处理包括非线性配准和深度学习去颅骨,但未进行N4偏置校正。数据集以NIfTI格式提供,许可证为CC BY 4.0。
The UCSF-PDGM (University of California San Francisco Preoperative Diffuse Glioma MRI) BraTS-format structural subset is a cohort of preoperative 3T brain MRI from 495 patients with histopathologically confirmed diffuse gliomas. It provides four structural MRI modalities (T1, T1c, T2, FLAIR) co-registered to the 240×240×155 BraTS grid (1 mm isotropic, skull-stripped) and expert-corrected tumor segmentation masks per patient. The dataset includes clinical metadata such as sex, age, WHO CNS grade, WHO-2021 diagnosis, IDH status, MGMT status, overall survival, and more. Tumor segmentation follows BraTS labeling conventions: 0 for background, 1 for non-enhancing tumor core/necrosis, 2 for peritumoral edema, and 4 for enhancing tumor. Note: Approximately 60% of cases overlap with BraTS 2021/2023, requiring exclusion of overlapping cases for independent benchmarking to avoid train/test leakage. Preprocessing involves co-registration and skull-stripping, but no N4 bias correction. Data is provided in NIfTI format under the CC BY 4.0 license.




