Artificial intelligence in dermatology: а scoping review

Abstract

Due to the large volume of diverse data regularly received, automation of routine processes in dermatology is a highly relevant task. Artificial intelligence (AI) may provide effective solutions for automating various processes in dermatology. Review Aim: To assess the current state of development and implementation of AI in dermatology and identify key challenges hindering AI integration into clinical practice. A literature search was conducted in PubMed and the Russian Science Citation Index (RSCI) databases, as well as in the Federal Service for Surveillance in Healthcare (Roszdravnadzor) register, to identify registered medical devices incorporating AI. The time frame covered 2019 to 2025. Bibliometric data, research focus, and type of pathology studied, the main methodological characteristics, the diagnostic accuracy of AI and medical staff, the number and experience of medical staff involved, and proven results of AI implementation were extracted from the articles. For the assessment of bias risk, the QUADAS-CAD was used. A total of 41 out of 270 identified references were included in the systematic review. Most studies focused on diagnosing malignant skin neoplasms (65.85%), melanoma (51.22%). In the analyzed studies, AI demonstrated high diagnostic performance comparable to those of experienced medical specialists. Median value (n = 27) for accuracy of neural networks in diagnosing malignant skin neoplasms was 80% (95% CI: 76.55–83.45%).Of the algorithms analyzed, eight have the status of medical devices with AI, and four are mobile applications that can be used to diagnose skin diseases. AI implementation in dermatology is at an advanced stage, with 19.5% of studies analyzed reaching commercial deployment and product distribution levels. However, further research is needed in this area, with improvements in the quality of methodologies used to assess the diagnostic accuracy of AI.

Ähnliche Arbeiten