ZebraMap: A Multimodal Structured Dataset for Rare Disease Cases

ZebraMap is a multimodal knowledge map designed to consolidate information on rare diseases and convert narrative case evidence into structured, machine-readable data. Utilizing Orphanet as the disease registry, we identified 1,727 rare diseases and established connections to PubMed case reports. A total of 47,960 full-text case report articles were retrieved that satisfied predefined inclusion criteria. Publication metadata, patient demographics, clinical narratives, and associated images were extracted. The final dataset comprises 74,705 structured patient-level case texts and 105,124 clinical images, all of which are associated with specific patient IDs, disease entries, and publications.