DermArtifactDB: a multi-label dataset for artifact annotation in public dermoscopic image repositories

Dermoscopic images acquired in routine clinical practice frequently contain visual artifacts—such as hair, ruler markings, gel-related structures, or ink traces—that may obscure diagnostically relevant patterns and affect the performance of computer-aided diagnosis systems. Despite their relevance, publicly available dermoscopic datasets rarely provide explicit annotations describing the presence and type of such artifacts. DermArtifactDB is a standardized multi-label dataset of dermoscopic artifact annotations spanning three widely used public repositories: ISIC-2019, PH2, and Derm7pt. The dataset provides binary annotations for seven common artifact categories (hair, dark corners, gel border, gel bubbles, ruler, ink, and patches), reflecting realistic acquisition conditions encountered in clinical practice. Artifact labels were generated using a scalable hybrid annotation framework that combines classical image-processing detectors, Transformer-based multi-label validation, and expert manual review to ensure annotation consistency and reliability across datasets. In addition to per-image annotations, the repository includes dataset-level summary statistics describing artifact prevalence, multi-artifact distributions, and representative co-occurrence patterns. The repository also provides optional attention-based visualization images highlighting regions associated with artifact predictions, intended solely for qualitative inspection and interpretability purposes. By enriching existing dermoscopic repositories with explicit artifact metadata, DermArtifactDB supports systematic studies of artifact-related bias and facilitates the development of artifact-aware preprocessing and learning pipelines. This resource contributes to more transparent, interpretable, and robust dermoscopic image analysis.