LH2-data-labs/longitudinal-dermatology-atlas

大多数皮肤病AI数据集是横断面快照，主要代表Fitzpatrick I–III皮肤类型。本数据集颠覆了这两个限制：它提供了**纵向临床记录**，具有多年每患者随访（成熟队列5年以上），并且完全基于**南亚人群**（Fitzpatrick IV–VI），解决了皮肤病AI中众所周知的皮肤类型偏见。此样本包含10名去标识化患者跨越197次临床就诊的结构化就诊级别元数据，展示了纵向模式。完整数据集包括60,000多条纵向记录，且正在持续招募中。**此样本仅包含元数据**。临床摄影（每次就诊每个解剖区域300多张高分辨率图像）在签署数据使用协议后可通过门控访问获得。

Most dermatology AI datasets are cross-sectional snapshots predominantly representing Fitzpatrick I–III skin types. This dataset inverts both limitations: it provides **longitudinal clinical records** with multi-year per-patient follow-up (5+ years for mature cohorts) and is anchored entirely in **South Asian populations** (Fitzpatrick IV–VI), addressing the well-documented skin-type bias in dermatology AI. This sample contains structured visit-level metadata for 10 de-identified patients across 197 clinical visits, demonstrating the longitudinal schema. The complete dataset includes 60,000+ longitudinal records with active enrolment ongoing. **This sample is metadata-only.** Clinical photography (300+ high-resolution images per patient across anatomical regions per visit) is available under gated access with a signed Data Use Agreement.