Applications of artificial intelligence in rare skin diseases and skin cancers: A scoping review

Background Rare skin diseases and rare skin cancers are individually uncommon but collectively associated with substantial diagnostic, prognostic, and healthcare-access challenges. Their heterogeneous presentations, limited specialist availability, and small patient populations make them difficult to study and manage. Although artificial intelligence (AI) is increasingly used in dermatology, its application to rare dermatologic conditions remains limited by small and fragmented datasets, restricted data sharing, limited multimodal integration, inconsistent validation, and insufficient reporting of demographic and fairness-related variables. Objective This scoping review aimed to map how AI has been applied to rare skin diseases and rare skin cancers and to identify methodological and translational gaps across data sources, preprocessing practices, model types, validation strategies, performance reporting, and demographic and clinical-variable reporting. Methods We conducted a scoping review across eight databases: Ovid MEDLINE, Embase, CINAHL, IEEE Xplore, ACM Digital Library, Scopus, Web of Science, and Google Scholar. The final search was completed on January 28, 2025, with biweekly updates through April 28, 2025. Empirical studies applying AI to rare skin diseases or rare skin cancers recognized by Orphanet and/or the National Organization for Rare Disorders were included. Two reviewers independently screened records, with disagreements resolved by a third reviewer. Inter-reviewer agreement was substantial to almost perfect (Cohen's κ = 0.78–0.83). Data were extracted on study design, disease category, data modality, preprocessing, AI methods, validation strategies, performance metrics, and demographic and clinical reporting, and were synthesized descriptively. Results A total of 68 studies published between 2007 and 2025 were included. Most studies were retrospective (49/68, 72.1%) and conducted in clinical settings (46/68, 67.6%), with more than half published since 2023. Research was concentrated on rare skin cancers and cutaneous lymphomas, followed by connective tissue and autoimmune inflammatory diseases. Imaging and tabular data were the most common data formats, whereas multimodal datasets were uncommon. Most studies relied on small, closed, or institution-specific datasets. Detection tasks predominated (44/68, 64.7%), while predictive modeling, including prognosis and treatment-response prediction, was less common (24/68, 35.3%). Traditional machine-learning methods were used more frequently than deep-learning approaches (35/68, 51.5% vs 23/68, 33.8%). Reported algorithm performance was generally encouraging, particularly for image-based detection and classification; however, evaluation was largely internal. Cross-validation was common (44/68, 64.7%), whereas external validation was reported in only 9/68 studies (13.2%). Reporting of demographic and clinical variables was limited: 18 of 68 studies reported participant age (26.5%), 13 reported sex or gender (19.1%), 3 reported ethnicity or race (4.4%), and 16 reported clinical characteristics (23.5%). Conclusion AI research in rare dermatology is expanding rapidly and shows technical promise, especially for image-based detection and classification. However, the field remains at an early translational stage, with evidence constrained by small and non-public datasets, limited multimodal integration, inconsistent preprocessing and reporting, and infrequent external validation. Incomplete demographic and clinical reporting further limits assessment of generalizability, fairness, and clinical applicability. This review provides a unified evidence-maturity framework and identifies key methodological priorities for developing clinically robust, reproducible, and equitable AI systems for rare skin diseases and rare skin cancers.