Beyond the Visible Lesion: Evaluating Dermoscopic Data and Breslow Thickness in Cutaneous Melanoma—A Retrospective Cross-Sectional Study

Abstract

Background: Dermoscopy provides morphological information for the assessment of cutaneous melanoma. However, the reliability of secondary image-based research depends on accurate lesion identification, consistent acquisition characteristics, and the availability of relevant clinical and histopathological metadata.
Objective: To evaluate the distribution of source-reported Breslow thickness, completeness of dermoscopic and clinical metadata, lesion identification, and feasibility of investigating dermoscopic feature–thickness associations using the Early-Stage Melanoma Benchmark (EMB).
Methods: This retrospective cross-sectional descriptive study analyzed a fixed EMB dataset retrieved on October 22, 2026. Image integrity, acquisition modality, thickness labels, lesion identifiers, and metadata completeness were assessed. Dermoscopic records with positive source-reported Breslow thickness were summarized by source. Records with explicit lesion identifiers were grouped separately. No missing clinical information or dermoscopic feature scores were imputed.
Results: The dataset contained 1,104 images, including 1,017 dermoscopic records. Of 438 dermoscopic records with positive thickness values, median Breslow thickness was 0.50 mm (interquartile range [IQR], 0.30–0.90 mm); 334 (76.3%) had values ≤1.0 mm. ISIC and Dermoscopy Atlas contributed 339 and 99 positive-thickness records, respectively, with medians of 0.40 and 0.70 mm. Explicit identifiers were available for 67 records corresponding to 58 source-defined lesions, of which 57 (98.3%) had values ≤1.0 mm. Polarized acquisition was documented in 208/438 records (47.5%). Dermoscopic feature annotations were unavailable, precluding association analyses.
Conclusion: EMB supports description of the available thickness spectrum and evaluation of methodological feasibility. However, incomplete lesion linkage, uneven thickness representation among identifiable lesions, and missing feature annotations limit its current suitability for lesion-level association analyses. Source reconciliation and standardized expert annotation are required before examining relationships between dermoscopic features and Breslow thickness.