AI and Imaging Technologies in Early Melanoma Detection: Balancing Sensitivity, Overdiagnosis, and Access

Conclusions

• AI and related technologies may help dermatologists and other clinicians triage pigmented lesions better, but they are not likely to replace expert clinical judgment or eliminate diagnostic uncertainty.
• The main value of these tools may be in reducing unnecessary biopsies and helping identify which lesions are truly concerning, rather than simply finding more melanomas.
• Trying to maximize sensitivity by labeling melanoma in situ and severely dysplastic nevi as positives can inflate overdiagnosis and reduce specificity, so endpoints need to be chosen carefully.
• Even intensive screening with photography, dermoscopy, and serial imaging does not fully prevent thicker melanomas from being found, suggesting technology cannot fully screen us out of melanoma mortality.
• Tools that work well as rule-out tests may be more useful in low-pretest-probability settings, while their impact depends heavily on who uses them and in what clinical context.
• Real-world performance can differ from trial performance, and the usefulness of a test depends on how it changes management, referral patterns, and biopsy decisions.
• Some lesion-level devices show high sensitivity but low specificity, which can increase biopsies if thresholds are not set thoughtfully.
• The DermTech genomic assay and other molecular tests appear most useful when used to help defer biopsy in selected lesions rather than as broad screening tools.
• Whole-body imaging and AI-based lesion tracking may be more promising for finding biologically important, new, or changing lesions while reducing biopsies of benign nevi.
• The strongest future role for melanoma technology may be improving access and triage, especially in settings with limited dermatology availability, while preserving clinician time and minimizing low-value care.