P003 Artificial intelligence in dermatology: current clinical applications, limitations and implementation challenges

Abstract

Artificial intelligence (AI) is now actively influencing skin cancer triage, dermatopathology slide review and clinical documentation in dermatology. As these tools enter routine practice, dermatologists are increasingly required to interpret, supervise and integrate AI outputs into clinical decision making. This review examines where AI has demonstrated meaningful clinical progress in dermatology, where limitations persist, and why these distinctions matter for practising dermatologists. Using a narrative review, we synthesized evidence on clinically relevant AI applications, real-world performance and regulatory considerations across major domains of dermatologic practice. Advances have occurred in three primary domains. Firstly, AI-enabled diagnostic and triage systems for skin cancer have demonstrated high sensitivity in controlled studies and selected real-world settings, supporting lesion prioritization and referral management. Secondly, in dermatopathology, deep learning models applied to digitized slides can accurately detect common tumours and highlight areas of concern, with growing evidence supporting their role as diagnostic aids within defined workflows. Thirdly, nondiagnostic applications, particularly AI-assisted documentation and ambient scribing, represent one of the most rapidly adopted and immediately impactful uses of AI, with clear potential to reduce documentation burden and improve clinical efficiency. Despite these advances, several practical considerations continue to shape real-world implementation. While many AI systems demonstrate strong performance in defined use cases, effectiveness can vary across settings, populations and workflows. Ongoing needs include broader external validation, continued evaluation across diverse skin types, clear regulatory pathways, and thoughtful integration into clinical practice that supports, rather than complicates, dermatologist decision making. In conclusion, this review provides a practical framework for understanding where AI currently adds value in dermatology, where caution is warranted, and how dermatologists can guide responsible implementation. By distinguishing clinically mature applications from higher-risk use cases, this work aims to support informed adoption of AI tools that enhance, rather than replace, clinical judgement.