AI in Melanoma Diagnosis: Ongoing Controversies and New Insights

Conclusions

• AI tools for melanoma diagnosis should prioritize the detection of invasive melanomas over melanoma in situ (MMIS) to maximize clinical utility.
• Training AI systems effectively involves using invasive melanomas as malignant cases while potentially considering dysplastic nevi as benign.
• The FDA has specific criteria that influence how AI tools should be validated and what lesions they need to detect, impacting the design of these technologies.
• A balance must be struck between sensitivity and specificity in AI tools: high sensitivity may lead to many false positives, while optimizing for specificity may miss some cases of MMIS.
• In real-world applications, the specificity of AI tools often suffers, which necessitates careful consideration of biopsy referral rates and the potential for patient anxiety over false positive results.
• Regression of AI tools and their respective accuracy metrics demonstrate varied performance depending on the clinical context and user experience, indicating the importance of user training.
• Innovative technologies like Nevisense and the use of telemedicine with AI support have shown promise in enhancing melanoma detection rates and potentially reducing the burden of unnecessary biopsies.
• Future developments in AI for dermatology should focus on improving specificity without sacrificing sensitivity, facilitating better clinical workflows.
• Clinical trials demonstrate significant performance with various AI tools, suggesting their potential as adjuncts in primary care settings for melanoma screening.