The performance of automated diagnostic support systems depends not only on the volume of images, but also on the reproducibility of the image acquisition procedure, the completeness of structured metadata, and the reliability of diagnostic labels. Unfortunately, international collections were primarily created under conditions that differ substantially from routine Russian outpatient practice and mobile dermatoscopy. Therefore, the training base is insufficient and an independent (external) evaluation set is required to test models trained on open collections.
This study presents a methodology for constructing a clinically verified dataset of dermatoscopic images for medical informatics research, together with its software implementation. The proposed methodology integrates three interconnected components:
(1) a standard operating procedure (SOP) for acquiring images via mobile dermatoscopy,
(2) an information model comprising 16 structured metadata fields organized into six clinically oriented blocks in ISIC-compatible notation,
(3) a multi-stage expert verification of diagnostic labels (initial clinical annotation, consensus review by three specialists, and histological confirmation of all malignant neoplasms)
the software implementation couples a mobile client application, a relational store, the Colba collaborative annotation environment, and a de-identified export module.
Using this methodology, a dataset of 1 026 unique dermatoscopic images from 443 patients was collected between June 2025 and May 2026. At acquisition, 32 technically unusable frames were rejected under the SOP criteria and never entered processing; of the 1 044 records that did, 18 duplicates were excluded. The dataset includes nine nosological categories; all 39 malignant lesions (18 melanomas, 15 basal cell carcinomas, and 6 squamous cell carcinomas) were histologically verified. Patient age ranged from 2 to 90 years (median 38), with 279 females (63 %) and 164 males (37 %).
To support work under pronounced class imbalance, four consistent levels of diagnostic-category aggregation are defined (from binary to nine-class), and the co-occurrence of nosological categories within a patient record is analysed:107 of 443 patients present two or more distinct categories. Each image is accompanied by expert-annotated dermatoscopic structures and an explicit verification_stage field indicating the level of diagnostic confirmation.
The presented formalization of a reproducible approach to preparing clinical dermoscopy datasets, along with the description of the supporting software tools, is applicable to the creation of resources for the external validation of clinical decision support algorithms in dermato-oncology. (In Russian).