AI that can predict race from x-ray images baffles researchers


The inexplicable ability of an AI program to predict a patient’s stroke with 90% accuracy simply by reading x-rays has researchers at MIT and Harvard scratching their heads.

The findings raise concerns about racial bias and the ability of AI to diagnose a patient based on their race rather than their individual needs. “These findings demonstrate that the medical technology we use can capture information that we didn’t know was possible or unintended,” said Marzyeh Ghassemi, study co-author and assistant professor of electrical engineering and science. computer science at MIT, at HCB News. “This information can then be used by a large-capacity model in ways that we don’t realize. Any care customization may not be desirable or appropriate for models trained on large amounts of data from processes biased humans”

The researchers trained a deep learning model to identify race from X-rays of the patients’ chest, head, and spine that provided racial information that was not included in the X-ray itself. same. In several tests, they evaluated its capabilities based on variable differences such as anatomy, bone density, and image resolution. For example, with bone density, they speculated that since black patients generally have higher bone mineral density, color differences of the thickest and thinnest parts of the bone allowed the model to identify race. , but even when they applied a filter to the images, the model was still able to accurately predict race.

Advertising statistics

Hours posted: 26803
Number of visits: 221

Although scientists are still unsure how it could do this, Ghassemi theorizes that it could be possible that X-ray and CT scanners are able to detect a higher content of melanin – the pigment that determines skin color – of darker skin and incorporate that information into the digital. pictures. She told the Boston Globe that it’s possible human radiologists have never noticed it before, but stresses that more research needs to be done.

“There’s a lot of good research showing racial bias in diagnostic and treatment plans without any machine learning models – it’s a problem in any system with human judgments,” she told HCB. News. “By engaging with technology, we want to make sure it doesn’t get worse. Systemic audits and strong regulatory guidelines, coupled with better support and training for clinical staff are all potential ways to address this risk.

The findings were published in Digital Health The Lancet.

Back to HCB news

Previous Check out these 37 haunting images from inside Staten Island's spooky Kreischer Mansion
Next Ex-Red Arrow found with stash of child abuse footage avoids jail