AI analysis of fundus photographs has potential for racial bias
Click Here to Manage Email Alerts
Key takeaways:
- AI models have the potential to learn features associated with race and infer race from retinal images.
- This increases the risk for racially biased performance in diagnostic tasks.
AI algorithms have the potential to infer race not only from color retinal fundus images but also from grayscale retinal vessel maps that eliminate any possibility for humans to infer race, according to a study.
The authors expressed concern about potential AI racial biases that could ultimately affect diagnosis and treatment decisions.
Image: Adobe Stock
The study included a selected sample from the Imaging and Informatics in Retinopathy of Prematurity cohort study, including 94 infants reported as Black and 151 infants reported as White by their parents/guardians. A total of 4,095 color retinal fundus images were collected to train, validate and test 40 ResNet-18 convolutional neural network (CNN) models. All color retinal fundus images were then segmented into grayscale retinal vessel maps, which were, in turn, iteratively transformed via thresholding, binarizing or skeletonizing to eliminate any potential information on vessel pigmentation, size and caliber that could differentiate between Black and White race.
The models had near-perfect ability to predict self-reported race (SRR) from color retinal fundus images.
“Although race itself is a social construct, it is associated with variations in skin and retinal pigmentation,” the authors wrote.
However, the CNN was able to infer Black vs. White race with comparable precision from grayscale retinal vessel maps, which is something human readers cannot do. “Even images that appeared devoid of information to the naked eye retained predictive information,” the authors wrote.
“Results of this diagnostic study suggest that it can be very challenging to remove information relevant to SRR from fundus photographs. As a result, AI algorithms trained on fundus photographs have the potential for biased performance in practice,” they wrote.
Collapse
Read more about
Click Here to Manage Email Alerts