Artificial intelligence valuable in skin cancer detection
Researchers have shown artificial intelligence may one day automate melanoma diagnosis, with an international study revealing a fast-learning computer can outperform experienced dermatologists at detecting the deadly cancer.
In the unique investigation, a ‘deep learning convolutional neural network’ (CNN) was shown more than 100,000 images of malignant melanomas and benign moles in order to ‘teach it’ how to identify the aggressive cancer.
The researchers from Germany, the United States and France then engaged 58 dermatologists from around the world – with varying experience – to compete against the artificial intelligence.
The dermatologists had to review and diagnose malignant melanoma or benign moles – first using images alone and then, again, supported by patient histories. The computer was more accurate.
Dr Victoria Mar from The Alfred’s Victorian Melanoma Service and Monash University is an experienced dermatologist – and was invited to write an editorial on the future role of artificial intelligence (AI) for melanoma diagnosis.
“The international study showed that, on average, dermatologists accurately detected almost 87 per cent of skin cancers from the images, compared to 95 per cent for the neural network,” Dr Mar said.
Dr Mar said the expert dermatologists performed better than the less experienced dermatologists at detecting malignant melanomas.
“Diagnostic accuracy for melanoma is dependent on the experience and training of the treating doctor, and AI can offer a more standardised level of diagnostic accuracy.”
“The findings show that AI is capable of out-performing dermatologists in the task of melanoma detection and, integrated as an additional diagnostic aide, could assist clinicians to reach an appropriate management decision.
“There is currently no substitute for a thorough clinical examination and, while automated diagnosis may eventually change the way we diagnose skin cancers, there is much more work to be done to implement this exciting technology safely into routine clinical care.”
The paper, Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists appears in the current issue of the Annals of Oncology.
