AI tool can diagnose skin cancer 99% correctly

The research was presented at the European Congress of Dermatology and Venereology 2023 on October 11. According to experts, AI can detect potentially fatal cancer cases at an early stage.

“The study demonstrates that AI is improving and learning rapidly, with high accuracy. The foundation is improvements in AI training techniques and the quality of data used to create AI,” said Dr Kashini Andrew, a specialist at University Hospitals Birmingham NHS Foundation Trust, lead author of the study.

The new AI software evaluated more than 22,300 suspected skin cancers over 2.5 years. It detected 189 of 190 skin cancers (99.5%), 541 of 585 precancerous lesions (92.5%), and all 59 melanomas.

The first AI model, tested in 2021, detected only 85.9% (195 of 227) of melanoma cases, 83.8% (903 of 1,078) of skin cancers, and 54.1% (496 of 917) of precancerous lesions.

According to Dr. Andrew, the latest version of the software recorded more than 1,000 live consultation calls at the care agency from April 2022 to January 2023, supporting many patients in need of urgent care. However, the researchers note that the software cannot replace a dermatologist and should be used in parallel with professional medical services.

“We want to emphasise that AI should not be used as a standalone tool for detecting skin cancer. It cannot replace a dermatologist,” said Dr Irshad Zaki, consultant dermatologist at UHB NHS Foundation Trust, who co-authored the study.

During testing, the software missed one case of basal cell carcinoma, which was later detected by a “safety net” of dermatologists, further demonstrating that AI should not be the sole diagnostic tool.

“With further research and appropriate monitoring, AI could be deployed as a triage tool. However, any approach would need to be proven cost-effective. AI is not currently a stand-alone tool in dermatology,” Dr Andrew added.