Answers are grounded in this paper and cite their sources.
The United States faces a critical dermatology workforce shortage, with more than 60% of counties lacking specialist access. Primary care providers (PCPs) are often tasked with managing suspicious skin lesions but demonstrate lower diagnostic accuracy than dermatologists. Artificial intelligence (AI) tools offer potential solutions, and meta-analytic evidence demonstrates that AI assistance improves PCP sensitivity by 13 percentage points and specificity by 11 percentage points. Yet only 1 Food and Drug Administration–authorized device (January 2024) is cleared for nondermatologist use in the United States. This device achieves 96% sensitivity but only 21% specificity, meaning most positive results are false alarms. A separate concern is that AI training data sets systematically under-represent darker skin tones: only 2.1% of images in public data sets include Fitzpatrick skin type data and algorithm performance degrades substantially on Fitzpatrick types V to VI. The convergence of the first Food and Drug Administration authorization for primary care use with new evidence documenting algorithmic bias across skin tones creates an urgent need for a practical, evidence-based review. This literature review examines the current evidence, available devices, training data limitations, and practical implementation considerations for PCPs considering AI-assisted skin cancer detection.
No takes yet. Share an insight, caveat, or question.
Hensley et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: