BIMCV-Prostate-Dataset V1

⚠️ Data Availability Notice The BIMCV Prostate Dataset is currently subject to a regulatory reassessment process conducted by the competent authorities and collaborating healthcare institutions in the Valencian Community (Spain). This process aims to ensure full compliance with updated legal and ethical frameworks governing the sharing of medical data. Consequently, although the dataset is publicly registered and associated with a peer-reviewed publication, access to the data is temporarily suspended until the review process is completed and all requirements are fulfilled. Access will be reinstated as soon as authorization is granted under the revised regulations. We remain committed to transparency and to enabling data access in accordance with applicable standards. ---The BIMCV Prostate Dataset is a comprehensive and diverse dataset that includes a total of 9,341 prostate MRI sessions, distributed among 8,441 subjects, collected from 16 healthcare centers in the Valencian Community, Spain. This dataset is structured according to the MIDS (Medical Imaging Data Structure) standard, ensuring consistent and accessible organization for researchers, facilitating data use and analysis. The first version of the dataset focuses on sessions that contain the three mentioned imaging modalities (T2W, DWI, and ADC), resulting in a total of 1,730 complete sessions, with a total of 4,663 samples for training, of which 2,594 are csPCa positive and 2,069 are csPCa negative. This information can be found in the table available on GitHub. The dataset includes MRI images in three modalities: T2-weighted images (T2W), diffusion-weighted images (DWI), and apparent diffusion coefficient (ADC) maps. In total, the dataset includes 32,662 T2W images (62.97%), 8,036 DWI images (15.49%), and 11,167 ADC maps (21.53%), including both the original maps and those calculated from the available DWI images. This additional calculation process was carried out to ensure the dataset's integrity and consistency, allowing for comprehensive analysis in the field of prostate oncology. The exploratory data analysis (EDA) performed on this dataset has provided insights into the characteristics and distribution of the images, ensuring the dataset's representativeness and diversity. For example, it was found that Health Center 5 contributed the highest proportion of sessions (15.6%), followed by Health Center 7 (12.3%) and Health Center 17 (10.5%). This level of diversity in data sources ensures that the dataset encompasses a wide range of imaging acquisition practices and patient demographics, improving the generalization of artificial intelligence models developed with this data. Additionally, the analysis of the distribution by MRI equipment manufacturer revealed that most images were acquired with General Electric equipment (66.7%), followed by Philips (25.1%) and Siemens (8.13%). Similarly, most sessions were conducted with 1.5 Tesla machines (63%), followed by 3.0 Tesla machines (36.5%), reflecting standard clinical practices in the region. Regarding the distribution of labels within the dataset, of the total cases, 4,871 (approximately 52%) are labeled as csPCa positive, while 3,514 cases (approximately 37%) are labeled as csPCa negative.⚠️ 数据可用性声明 BIMCV前列腺数据集目前正处于西班牙巴伦西亚自治区主管部门及合作医疗机构开展的监管重审流程中，该流程旨在确保数据集完全符合医疗数据共享相关的最新法律与伦理框架要求。 因此，尽管该数据集已完成公开注册并关联一篇同行评议论文，但其访问权限已临时暂停，直至重审流程完成且所有要求均得到满足。 待修订后的法规获批后，数据访问将立即恢复。我们始终秉持透明原则，并将严格遵循适用标准保障数据可及性。 --- BIMCV前列腺数据集是一个全面且多样化的数据集，共包含来自西班牙巴伦西亚自治区16家医疗中心的9341次前列腺磁共振成像（MRI）扫描，涉及8441名受试者。该数据集遵循MIDS（医学成像数据结构，Medical Imaging Data Structure）标准进行组织，可为研究人员提供统一且易于访问的数据结构，便于开展数据使用与分析工作。 该数据集的首个版本聚焦于包含上述三种成像模态（T2加权成像（T2W）、扩散加权成像（DWI）及表观扩散系数（ADC）图谱）的扫描，共计1730次完整扫描，其中4663个样本用于模型训练，包括2594例临床显著性前列腺癌（clinically significant prostate cancer，csPCa）阳性样本与2069例阴性样本。相关信息可在GitHub提供的表格中查询。 数据集包含三种模态的MRI图像：T2加权图像（T2W）、扩散加权图像（DWI）以及表观扩散系数（ADC）图谱。数据集总计包含32662张T2W图像（占比62.97%）、8036张DWI图像（占比15.49%）及11167幅ADC图谱（占比21.53%），其中既包含原始生成的ADC图谱，也包含由现有DWI图像计算得到的图谱。开展此项额外计算流程旨在保障数据集的完整性与一致性，可为前列腺肿瘤学领域的全面分析提供支撑。 针对该数据集开展的探索性数据分析（Exploratory Data Analysis，EDA）已揭示了图像的特征与分布情况，确保了数据集的代表性与多样性。例如，分析显示5号医疗中心贡献的扫描占比最高（15.6%），其次为7号医疗中心（12.3%）与17号医疗中心（10.5%）。这种多源数据的多样性确保数据集覆盖了广泛的成像采集实践与患者人口统计学特征，可提升基于该数据集开发的人工智能模型的泛化能力。 此外，针对MRI设备制造商的分布分析显示，大部分图像由通用电气（General Electric）设备采集（占比66.7%），其次为飞利浦（Philips，占比25.1%）与西门子（Siemens，占比8.13%）。同理，大部分扫描使用1.5特斯拉设备完成（占比63%），其次为3.0特斯拉设备（占比36.5%），这与该地区的标准临床实践相符。 关于数据集中的标签分布：在全部病例中，4871例（约占52%）被标记为临床显著性前列腺癌阳性，3514例（约占37%）被标记为阴性。