AI skin cancer diagnoses risk being less accurate for dark skin – study – The Guardian

AI systems being developed to diagnose skin cancer run the risk of being less accurate for people with dark skin, research suggests.

The potential of AI has led to developments in healthcare, with some studies suggesting image recognition technology based on machine learning algorithms can classify skin cancers as successfully as human experts.

NHS trusts have begun exploring AI to help dermatologists triage patients with skin lesions.

But researchers say more needs to be done to ensure the technology benefits all patients, after finding that few freely available image databases that could be used to develop or “train” AI systems for skin cancer diagnosis contain information on ethnicity or skin type. Those that do have very few images of people with dark skin.

Dr David Wen, first author of the study from the University of Oxford, said: “You could have a situation where the regulatory authorities say that because this algorithm has only been trained on images in fair-skinned people, you’re only allowed to use it for fair-skinned individuals, and therefore that could lead to certain populations being excluded from algorithms that are approved for clinical use.

“Alternatively, if the regulators are a bit more relaxed and say: ‘OK, you can use it [on all patients]’, the algorithms may not perform as accurately on populations who don’t have that many images involved in training.”

That could bring other problems including risking avoidable surgery, missing treatable cancers and causing unnecessary anxiety, the team said.

Writing in the journal Lancet Digital Health, Wen and colleagues report how they identified 21 open-access datasets for skin cancer images of which 14 recorded their country of origin. Of these, 11 included images only from Europe, North America and Oceania.

Few of the 21 datasets recorded the ethnicity or skin type of the individuals photographed, with the team noting that means it is unclear how generalisable algorithms based on them would be.

The team found just 2,436 of a total of 106,950 images within the 21 databases had skin type recorded. Of these, only 10 images were from people recorded as having brown skin and one was from an individual recorded as having dark brown or black skin.

Only 1,585 images contained data on ethnicity instead of, or as well as, information on skin type. “No images were from individuals with an African, African-Caribbean or South Asian background,” the team report.

“Coupled with the geographical origins of datasets, there was massive under-representation of skin lesion images from darker-skinned populations,” they add.

Wen said the omissions are unlikely to be deliberate but that there is a need for standards to ensure important information, including ethnicity or skin type, is reported with images. The authors add datasets used to develop AI systems should represent the populations the technology will be used in.

Charlotte Proby, professor of dermatology at the University of Dundee and British Skin Foundation spokesperson – who was not involved in the work – said the findings are of concern.