Expert Dermatologists Outperform AI Models in Real-World Skin Cancer Diagnosis

A comprehensive international diagnostic study confirms that expert dermatologists maintain superior accuracy over artificial intelligence in real-world skin cancer detection. While modern AI models outperform junior clinicians, they cannot match the diagnostic precision of specialists with over a decade of experience.

Researchers evaluated three AI systems alongside 652 physicians using 1,117 diverse skin lesion cases. Expert dermatologists correctly classified lesions in 74.2% of cases. The top-performing PanDerm unimodal model achieved 72.2% accuracy, surpassing physicians with fewer than three years of experience but failing to exceed senior specialist performance.

The findings highlight significant AI limitations outside controlled environments. Diagnostic accuracy declined when algorithms encountered atypical presentations common in routine practice. Consequently, researchers conclude AI serves best as a triage and support tool for junior staff rather than a replacement for expert clinical judgment.

Future dermatology workflows will likely integrate human-AI collaboration to reduce fatigue-related errors and assist less experienced practitioners. However, achieving the highest diagnostic standards in oncology remains dependent on seasoned medical expertise.