BRAINS-45K

该数据集名为BRAINS-45K，包含了44,756份来自公共来源的脑部MRI扫描数据，这些数据具有多样的采集参数和患者群体。该数据集融合了四个大型非标记数据集和五个挑战性数据集，旨在模拟临床数据的特点。数据集包含了1.5T和3T下采集的数据，并经过了预处理，包括转换到RAS坐标系统、重采样到等距的1毫米间隔，以及在体积层面上进行z轴标准化。使用该数据集需要遵守每个数据来源的个别法律要求。数据集的规模为44,756份扫描体积，任务类型为三维语义分割。

This dataset, designated as BRAINS-45K, consists of 44,756 brain MRI scan volumes sourced from public repositories, featuring diverse acquisition parameters and patient cohorts. It integrates four large-scale unlabeled datasets and five challenging datasets, aiming to mimic the characteristics of real-world clinical data. The data were acquired at both 1.5T and 3T magnetic field strengths, and have undergone preprocessing steps including conversion to the RAS coordinate system, resampling to isotropic 1-millimeter voxel spacing, and z-score normalization per scan volume. Usage of this dataset requires compliance with the individual legal requirements of each original data source. The dataset comprises a total of 44,756 scan volumes, and is tailored for 3D semantic segmentation tasks.

搜集汇总

背景与挑战

背景概述

BRAINS-45K是一个用于3D脑MRI预训练的大型数据集，与AMAES方法结合使用，旨在提升3D分割模型的性能。该数据集及其扩展版本（如FOMO60k和FOMO300K）已在Hugging Face平台上发布，供研究社区下载和使用。