Beyond diagnostic accuracy: evaluating the clinical utility of artificial intelligence for suspicious lesion triage

The recent systematic review and meta-analysis by Nirula et al. reports pooled sensitivity and specificity of 89% and 82%, respectively, for artificial intelligence (AI)-based differentiation between benign and malignant skin lesions. This highlights the diagnostic potential of these tools. However, diagnostic accuracy alone does not establish clinical utility; a tool that performs well in retrospective validation may not meaningfully alter referral patterns, reduce waiting times or improve healthcare efficiency in practice. As we move closer to the implementation of AI technologies in dermatology services, future studies should evaluate implementation outcomes - including referral avoidance, biopsy rates and healthcare utilization - alongside traditional accuracy metrics.