As the video highlights, the confusion which can arise in diagnosing conditions on dark skin could hold serious consequences.
Images of darker skin are historically underrepresented in the medical field. Conditions which can look quite different according to skin colour have largely been shown on white skin, ill equipping medical staff for their role.
This is thankfully beginning to change in the real and online world. Text books are becoming more inclusive, online libraries now offer images on varied skin colours and the ability to filter by them.
Future generations of doctors will have a broader view and a greater chance of recognising diseases, although not all recent ideas have kept up.
Building Bias In AI
Artificial intelligence systems are being developed to diagnose skin cancer. They may not have the capability of a good consultant but can be helpful, except where the lesions being studied are on brown, or black skin.
Research has found the error rate to be higher on darker skin, because the AI has the same issue as many doctors. Training images used to facilitate the AI’s knowledge have a significant bias towards lighter skin.
A study of over 100,000 medical database images used for AI training found that just 2% had the skin type recorded. Not surprising when there were only 11 images of people with brown, or black skin.
The Wider Outcome
Misdiagnosis is a potential outcome, along with issues in selecting the right treatment path. AI systems are now being developed to assist with this, to be used in selecting and prioritising clinical decisions.
For those of us who have easy access to doctors and dermatologists this could be an issue, more so if you do not. In many parts of the world, online triage and recommendations are seen as the best hope for medical care.
The people they are meant to serve may not have light skin but could be sent systems trained on this, errors seem inevitable.
Regulators have awkward decisions, do they only allow use of systems on patients which match the skin colour used for training. The alternative is to accept a better than nothing scenario, with inbuilt risk.
Medical training based on one skin colour is a recognised issue, as is AI bias. Camera software has been known to think people are blinking, simply because they have brown skin, or image recognition software fail to cope with black skin.
The risks in healthcare are serious, unnecessary anxiety for patients, missing skin cancers when they were easily treatable, or carrying out surgery which wasn’t needed.
No doctor will intend to create these risks and the position is improving. The point made in the video still bears thought and wherever you are, if you have a skin problem, try to see a dermatologist with experience across ethnic needs.
Dermatology for darker skin should be led by prompt diagnosis and treatment, based on knowledge of your individual skin type.