DermoPed: a paediatric dataset of non-dermoscopic skin images for dermatological AI research

DermoPed: A paediatric dataset of non-dermoscopic skin images DermoPed is a non-dermoscopic dermatology image dataset composed almost entirely of paediatric patients. It comprises more than 46,000 medium-to-high quality image crops from 229 distinct patients, covering nine dermatological conditions. Diseases. The dataset covers nine conditions: iatrogenic drug-induced exanthema, maculopapular exanthema, morbilliform exanthema, polymorphous exanthema, viral exanthema, urticaria, pediculosis, scabies, and chickenpox. Archive organisation. The dataset is distributed as nine ZIP archives, one per disease. Each archive unzips to a disease-named folder containing one subfolder per patient, named patient{n}. Per-patient structure. Each patient{n} folder contains: crops/ — the curated high-quality cropped images in PNG format (256×256), named {k}.png. masks/ — the corresponding binary masks in PNG format, named {k}_mask.png. challenging/ — lower-quality / non-curated crops and their masks, together with a quality_labels.json file reporting automatically computed image-quality labels for each such crop. metadata.json — the available metadata for that patient. metadata.json fields. Each file reports sex and access_age (the patient's age at the time of the clinical visit). Missing values are set to unknown. quality_labels.json. For each crop in the challenging subset, this file reports the raw image-quality measurements (mean luminance, Laplacian variance, and RMS contrast) and the labels assigned to it (underexposed, overexposed, blurred, low-contrast, or no-defect-detected), together with the thresholds used to derive them. The labels are produced automatically using standard no-reference image-quality measures and are intended as a fast, approximate, and reproducible annotation. Privacy and anonymisation. The dataset contains only de-identified skin crops. Crops containing recognisable identifying features (e.g. eyes, nose, mouth, or even partial fragments) were excluded, and the released metadata are limited to sex and age at access. No directly identifying information is included. Associated code. The preprocessing and labelling scripts are available at: https://github.com/aequitas-aod/DermoPed_automatic_preprocessing Funding. Supported by the European Union's Horizon Europe AEQUITAS research and innovation programme under grant agreement no. 101070363. Associated publication. A full description of the dataset is provided in the accompanying Data Descriptor (currently under review).