Artificial Intelligence in Dermatology: Current Applications and Future Directions
Sofía Pérez‐Lalinde, Francisco FloresABSTRACT
Background
Skin diseases affect billions globally, posing significant healthcare and financial burdens. Artificial Intelligence (AI) is rapidly restructuring medicine, and dermatology, as a highly visual, image‐dependent specialty, is uniquely positioned to integrate these technologies. However, the expanding literature on dermatological AI remains fragmented, presenting highly heterogeneous methodologies, localized datasets, and unaddressed ethical or regulatory challenges that complicate its translation into safe clinical practice.
Objectives
This scoping review aimed to systematically map the current applications of AI methodologies across dermatologic practice, identify the specific skin conditions evaluated, and describe the current limitations and evidence gaps preventing widespread clinical integration.
Methods
Following the Joanna Briggs Institute methodology and reported according to the PRISMA‐ScR guidelines, a systematic search was conducted across MEDLINE, Embase, and Scopus for articles published between January 2015 and September 2025, supplemented by registries (
Results
Of 584 records identified, 56 studies met the inclusion criteria. The evidence base was predominantly composed of secondary evidence syntheses (66%), followed by cross‐sectional studies (18%). Geographically, research was heavily clustered in high‐income nations (e.g., USA, Germany, China), with a critical underrepresentation of Latin America and Africa. Image‐based recognition and classification was the leading application (48%), primarily utilizing deep learning and convolutional neural networks for skin cancer detection (melanoma and non‐melanoma, 39%), achieving diagnostic metrics comparable to dermatologists in controlled benchmark settings. Other expanding domains included clinical decision support systems (21%), predictive analytics for inflammatory dermatoses (11%), teledermatology triage (11%), dermatopathology (5%), natural language processing, and multimodal vision‐language models. Key descriptive limitations identified across the literature include a reliance on retrospective convenience samples, a lack of external validation, dataset overlap, systematic bias regarding skin tone representation, and a complete absence of randomized clinical trials.
Conclusions
AI demonstrates robust capability in optimizing diagnostic workflows, triage, and lesion classification within dermatology. However, the current evidence base is largely exploratory, heterogeneous, and limited by demographic and geographic skew. Widespread clinical implementation remains premature. Future research must transition toward prospective clinical validation, standardized reporting, regulatory alignment, and inclusive dataset development to ensure equitable and safe deployment in global dermatologic care.