Artificial Intelligence Can Predict Race From X-Ray Images

Scientists worry that the accuracy of artificial intelligence could have a racial and biased impact.

REPUBLIKA.CO.ID, MASSACHUSETTS — New research has revealed a deep learning model (deep learning models) based on artificial intelligence (AI) can identify a person’s race only from X-rays. This was something that a human doctor could not possibly do.

These findings raise some troubling questions about the role of AI in medical diagnosis, assessment, and treatment: can racial bias be inadvertently applied by computer software when studying images like these?

Scientists trained the AI ​​using hundreds of thousands of existing X-ray images that were labeled with details of the patient’s race. Next, an international team of health researchers from the United States (US), Canada, and Taiwan tested their system on X-ray images that previous computer software did not have.

The AI ​​was able to predict the racial identity of the patients reported on these images with surprising accuracy. In fact, the data is quite accurate when the scans are taken from people of the same age and gender. The system reaches 90 percent level with multiple image groups.

“We aimed to carry out a comprehensive evaluation of the AI’s ability to recognize the racial identity of patients from medical images,” the researchers wrote in their published paper. SciencealertThursday (19/5/2022).

The study echoes the results of previous studies that found AI scans of X-ray images were more likely to miss signs of disease in black people. To stop that from happening, scientists need to understand why it happened.

AI mimics human thinking to quickly find patterns in data. Sscientists are currently unsure why AI systems are so good at identifying races from images that don’t contain that information, at least not on the surface.

Even when limited information is provided, by omitting clues about bone density for example or focusing on small parts of the body, the models still perform very well at guessing the races reported in the files.

It’s possible that the system discovered signs of melanin, the pigment that gives skin its color, that science doesn’t yet know.

“Our finding that AI can accurately predict self-reported race, even from distorted, clipped, and vocalized medical images, often when clinicians are unable to do so, creates a major risk for any application of the model in medical imaging,” write the researchers. .

This research adds to a growing body of evidence that AI systems can often reflect human biases and prejudices, whether they be racism, sexism or otherwise. There are still many unanswered questions from this research.

But right now it’s important to be aware of the potential racial biases that emerge in artificial intelligence systems—especially if we’re going to hand them more responsibility in the future.

“We need to take a break,” research scientist and physician Leo Anthony Celi of the Massachusetts Institute of Technology told Boston Globe.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Never miss any important news. Subscribe to our newsletter.