Bladder cancer (BCa), as the most common malignant tumor of the urinary system, has received significant attention in research on the clinical application of artificial intelligence algorithms. Nevertheless, it has been observed that certain investigations employ data from diverse medical facilities to train models for BCa, thereby posing a potential risk of leaking patients' privacy. Ensuring the privacy of patients during the training of machine learning algorithms is a vital consideration that deserves significant attention. Federated learning (FL) is an emerging machine learning paradigm that enables multiple entities to collaboratively build machine learning models while preserving data privacy and security. In this study, we present a multi-center BCa magnetic resonance imaging (MRI) dataset, aimed at evaluating the baseline performance of FL. The dataset comprises 275 three-dimensional bladder T2-weighted MRI scans collected from four medical centers, and each scan includes diagnostic pathological labels for muscle invasion and fine pixel-level annotations of tumor contours. Four FL methods are used to assess the baseline of the dataset for both the task of diagnosing muscle-invasive bladder cancer and automatic bladder tumor lesion segmentation.

The BCa dataset is provided under the terms of the Creative Commons Attribution 4.0 International (CC-BY-4.0) license (CC BY 4.0 Deed | Attribution 4.0 International | Creative Commons), aiming to promote open and collaborative research. Authors using this dataset for publications are respectfully requested to cite the original paper. We encourage contributors to openly share their methods and models, thereby enhancing reproducibility and advancing the field of medical imaging research.