AI risks racial bias, says study

A diagnostic study published in Jama Ophthalmology has identified potential racial bias in artificial intelligence (AI) algorithms using retinal fundus images.  

Researchers at the Singapore National Eye Centre collected 4095 retinal images and found that AI algorithms were able to learn racial biomarkers and interpret patterns of self-reported race (SRR) information, regardless of whether colour was removed or vessels were uniform width, due to variations in skin, retinal pigmentation and vessel differences between Black and White infants. 

It was found to be challenging to remove information relevant to SRR from retinal fundus images, resulting in risk of potential for biased performance in practice.  

The study concluded that understanding whether this information can be removed, without affecting the performance of AI algorithms, is critical in reducing the risk of racial bias in medical AI.