P005 The rise of artificial intelligence-powered teledermatology: transforming skin disease diagnosis, monitoring and personalized treatment plans in remote settings

Abstract

Teledermatology (TD) improves access to specialist assessment, but diagnostic confidence may be limited by variable image quality and rising demand. Artificial intelligence (AI), including computer vision and large language models, may augment TD through decision support, triage and longitudinal monitoring while maintaining clinician oversight. We aimed to synthesize evidence on AI-enabled TD for diagnosis, prioritization, monitoring and personalized remote management, and to identify implementation risks relevant to safe service delivery. A literature review was performed of English-language studies identified via MEDLINE, PubMed and related databases using terms including ‘teledermatology’, ‘artificial intelligence’, ‘deep learning’, ‘dermoscopy’, ‘triage’ and ‘monitoring’. We included real-time, store-and-forward and hybrid TD models. Evidence was grouped by use case (skin cancer, inflammatory dermatoses, workflow/triage, and patient-facing support). Across studies, AI demonstrated high sensitivity for melanoma and skin cancer detection, while specificity for benign and inflammatory conditions was more variable, contributing to false positives and avoidable downstream review. In studies of common dermatoses, reported diagnostic performance included ∼80% sensitivity with ∼50% specificity in some settings. For lesion classification, computer-vision approaches achieved ∼85% sensitivity and ∼90% specificity in selected datasets, with stronger performance in larger, more typical lesions. Beyond diagnosis, AI-assisted history taking and decision support improved data capture and may enhance asynchronous teledermatology. In comparative studies, an LLM-enabled workflow achieved ∼84% top-diagnosis concordance with clinician teleassessment in a case series. AI also supported monitoring and personalized management, including image-based severity scoring (e.g. psoriasis) with material improvement over average clinician scoring in one evaluation. Key implementation barriers included image-quality artefacts, under-representation of demographics and diseases in training datasets, privacy and cybersecurity risk, and lack of harmonized governance for clinician accountability and follow-up. AI-enabled teledermatology is most promising as a clinician-supervised decision support for triage and monitoring rather than autonomous diagnosis. Acceptance and safety will depend on representative datasets, standardized image capture, transparent performance reporting, and robust data governance integrated into real clinical pathways.