X-MAD: A Public Multi-Anatomy X-ray Dataset for Self-Supervised Learning in Low-Resource Radiographic Imaging

X-MAD is a public multi-anatomy X-ray dataset collected from Ibn Hayyan Specialist Hospital in Sana'a, Yemen. The dataset contains 5,595 de-identified JPG radiographic images covering heterogeneous anatomical regions, including the chest, skull, spine, abdomen, pelvis, and extremities. The dataset is released as an unlabeled image collection to support research in self-supervised learning, unsupervised learning, representation learning, clustering, image retrieval, masked image modeling, and annotation-efficient medical imaging. The dataset does not include diagnostic labels, anatomical labels, radiology reports, patient demographics, acquisition dates, or clinical metadata. All image files were renamed using anonymous sequential identifiers before release. The images were exported without patient-identifying information, and corrupted, black, or unreadable images were removed during quality control. X-MAD is intended for non-commercial research and educational use only. It is not intended for clinical diagnosis, treatment planning, direct patient care, or commercial use. The dataset is accompanied by documentation and benchmark code for self-supervised representation learning experiments.