Expanding the Brain Tumor Segmentation (BraTS) data to include African Populations

<p>The dataset is a collection of retrospective pre-operative brain magnetic resonance imaging (MRI) scans, clinically acquired from six diagnostic centers in Nigeria. The scans are from 146 patients who have brain MRIs indicating central nervous system neoplasms, diffuse glioma, low-grade glioma, or glioblastoma/high-grade glioma. The brain scans were multiparametric MR images (mpMRI), specifically T1, T1 CE, T2, and T2 FLAIR,  acquired on 1.5T MRI between January 2010 and December 2022. </p><p>Scans were obtained from different scanners using each center’s acquisition protocol. Each scan was de-identified and de-faced to remove personal identifiers and presented in their original state with respect to resolution and orientation. To ensure uniformity across scans and modalities, a standardized pre-processing protocol was applied to adjust the image dimensions and voxel sizes. The scans were extracted from the PACs as DICOM files and converted to the Neuroimaging Informatics Technology Initiative (NlfTI) file format to facilitate computational analysis, following the well-accepted pre-processing protocol of the International Brain Tumour Segmentation (BraTS) challenge. All scans were subjected to sanity checks to confirm the presence of all required sequences. Specifically, all mpMRI volumes were reoriented to the left posterior-superior (LPS) coordinate system, and the T1 CE scan of each patient was rigidly (6 degrees of freedom) registered and resampled to an isotropic resolution of 1 mm3 based on a common anatomical atlas, namely SRI. The remaining scans (i.e., T1, T2, FLAIR) of each patient were then rigidly co-registered to this resampled T1 CE scan by first obtaining the rigid transformation matrix to T1 CE, then combining with the transformation matrix from T1 CE to the SRI atlas, and resampling. The N4 bias field correction was applied in all scans to correct for intensity non-uniformities caused by the inhomogeneity of the scanner's magnetic field during image acquisition to facilitate an improved registration of all scans to the common anatomical atlas. Brain extraction was also performed using a standard process for  skull-stripping to remove all non-brain tissue (including neck, fat, eyeballs, and skull) from the image and create a brain mask to  enable further computational analyses. </p><h4>Inclusion Criteria</h4><p> All Brain MRI Scans of patients with clinical features of brain tumors from the study site acquired between January 2010 and December 2022, including, central nervous system (CNS) neoplasms, specifically diffuse glioma, or low-grade glioma (LGG) or glioblastoma/high-grade glioma (GBM/HGG).</p><h4>Exclusion Criteria</h4><p> Any brain image or scan that is not an MRI or acquired before January 2010 or after December 2022.</p><h4>Image Annotation</h4><p>The expert-annotated tumor sub-regions for each of the 146 cases are provided along with a metadata (csv file) of study location, scanner type, where available.</p><h4>Benefit to Researchers</h4><p>The contribution of BraTS-Africa dataset is two-fold: 1) its potential for use in research leading towards generalizable and inclusive diagnostic tools applicable across all settings including resource constrained environments, and 2) its ability to describe the peculiarities of neuroimaging in African settings.</p>

该数据集汇聚了来自尼日利亚六个诊断中心的回顾性术前脑磁共振成像（MRI）扫描图像，这些图像系临床采集。扫描对象为146名患者，其脑部MRI检查结果显示中枢神经系统肿瘤、弥漫性胶质瘤、低级别胶质瘤或胶质母细胞瘤/高级别胶质瘤。脑部扫描为多参数磁共振成像（mpMRI），具体包括T1、T1 CE、T2和T2 FLAIR序列，于1.5T MRI设备上采集，时间跨度为2010年1月至2022年12月。扫描图像均采用各中心的采集协议获取。每项扫描均进行去标识化和去面部处理，以去除个人信息标识，并以原始分辨率和方向呈现。为确保扫描和模态的统一性，应用了标准化预处理协议以调整图像尺寸和体素大小。扫描图像从PACS中提取为DICOM文件，并转换为神经影像学信息技术倡议（NIfTI）文件格式，以便进行计算分析，遵循国际脑肿瘤分割（BraTS）挑战赛中广为接受的预处理协议。所有扫描均经过完整性检查，以确认包含所有必需的序列。具体而言，所有mpMRI体积均重新定位至左侧后上（LPS）坐标系，每位患者的T1 CE扫描均进行刚性（6自由度）配准和重采样，以1 mm³的各向同性分辨率基于共同解剖图谱，即SRI进行。随后，每位患者的剩余扫描（即T1、T2、FLAIR）通过首先获得T1 CE的刚性变换矩阵，然后结合T1 CE到SRI图谱的变换矩阵进行重采样，与重采样的T1 CE扫描进行刚性配准。对所有扫描应用N4偏场校正，以纠正图像采集过程中由扫描器磁场不均匀性引起的强度不均匀性，从而促进所有扫描与共同解剖图谱的更好配准。此外，还采用标准的大脑剥离过程去除图像中的所有非脑组织（包括颈部、脂肪、眼球和颅骨），以创建大脑掩码，从而便于进一步的计算分析。