Study: AI Model Outperforms Early-Career Physicians in Skin Lesion Diagnosis

Data from a new study show that a foundational AI model outperformed early career dermatologists in skin lesion diagnosis, but not more experienced dermatologists, when using a dataset designed to reflect real-world clinical scenarios.

The study included 1,117 dermatologic cases collected between March 2023 and August 2025, incorporating clinical and dermoscopic images along with associated patient metadata. Investigators evaluated three AI systems: a first-generation convolutional neural network (CNN), a unimodal foundation model (PanDerm), and a multimodal foundation model. Human evaluators included 652 physicians with experience ranging from less than 1 year to more than 10 years in dermatology.

Diagnostic Accuracy of AI vs Human Readers in Real-World Dermatology

A total of 1,092 testing iterations were completed. Human readers achieved significantly higher multiclass diagnostic accuracy than the CNN (65.9% vs 56.7%; P < 0.001). The unimodal foundation model achieved a mean diagnostic accuracy of 72.2%, outperforming physicians with less than 3 years of experience (68.2%; P < .001) and matching the performance of clinicians with 3 to 10 years of experience.

Dermatologists with more than 10 years of experience achieved the highest overall diagnostic accuracy at 74.2%, outperforming all AI systems, including the unimodal foundation model (72.2%), multimodal model (66.3%), and CNN (56.7%).

"The modern foundation model surpassed readers with less than 3 years of experience on accuracy of skin lesion diagnosis and matched those with 3 to 10 years of experience but remained inferior to experts with more than 10 years of experience," the authors wrote. "These findings highlight both the promise and current limitations of AI in dermatologic diagnosis."

Source

Anriot J, et al. JAMA Dermatology. 2026. Doi:10.1001/jamadermatol.2026.1492