Similar skin cancer identification rates found between AI, dermatologists

The use of artificial intelligence to triage and identify certain skin cancers had results similar to dermatologists’ identifications, according to a study.

“For many patients, in-person access to a consultation with a board-certified dermatologist in a timely fashion can be difficult,” Mandy Majidian, BA, of Tulane University School of Medicine in New Orleans and Moy-Fincher-Chipps Facial Plastics & Dermatology in Beverly Hills, California, and colleagues wrote.

The COVID-19 pandemic made this access even more difficult for many, increasing the use of telemedicine in dermatology.

“Based on the prevalence and incidence of skin cancer, early detection and treatment is of critical clinical importance,” the authors continued.

Researchers studied how artificial intelligence (AI) software identified 100 clinical photographs of unbiopsied skin lesions, compared with evaluation from a three-person panel of board-certified dermatologists. The lesions were then biopsied for confirmation diagnoses.

The images included benign and malignant basal cell carcinomas (BCC), squamous cell carcinoma (SCC), seborrheic keratosis, malignant melanoma (MM) and actinic keratosis.

The researchers first considered lesion classification as benign vs. malignant. Malignancy was correctly identified by the AI software in 66% of the cases, whereas the dermatologists correctly identified malignancy 78.7% of the time (P < .05), which was a statistically significant difference; however, when only considering malignant lesions, the difference was not statistically significant.

Dermatologists saw an average 64.3% overall accuracy of diagnosis.

For the AI, the correct diagnosis as the top result occurred for only 38% of images. This accuracy increased to 63% when considering the top three differential diagnoses.

“Photographic identification of BCC, SCC and MM by AI as malignant did not significantly differ from the board-certified dermatologists,” the authors wrote. “With the increasing use of telemedicine, especially in light of a pandemic, the use of AI and a method of triaging patients with skin lesions is a useful option — especially when most skin cancers are first seen by non-dermatologists.”