AI Can Now Predict a Person's Race From Medical X-Rays

Source: Geralt/Pixabay

A key advantage of artificial intelligence (AI) machine learning is its ability to find patterns that are not apparent to the human eye. An astonishing new study published in The Lancet Digital Health shows that AI can predict a person’s race based on X-ray images with a high degree of accuracy, but scientists do not know exactly how.

“In this modelling study, which used both private and public datasets, we found that deep learning models can accurately predict the self-reported race of patients from medical images alone,” the global team of researchers from the United States, Canada, Taiwan, and Australia wrote. “This finding is striking as this task is generally not understood to be possible for human experts.”

To conduct the experiment, the researchers created AI deep learning models using three large, labeled datasets of chest X-rays to predict if the patient self-reported their racial identity as Black, Asian, or white. The images did not contain skin color, hair texture, or any other indicators of race. The team then evaluated the model with an unseen subset of the dataset and with data from a completely different dataset.

“The deep learning models assessed in this study showed a high ability to detect patient race using chest X-ray scans, with sustained performance on other modalities and strong external validations across datasets,” the researchers reported.

Then the team trained the deep learning algorithms with images from images that were not chest X-rays from various other body locations such as mammograms, lateral cervical spine radiographs, chest CTs, and digital radiography to determine if AI could detect race from imaging data beyond chest X-rays.

The AI deep learning algorithms were able to predict race more broadly across patient populations, clinical environments, and various medical imaging methods. The results suggest that AI is able to predict race beyond using chest X-ray images.

The researchers conducted a number of tests in an effort to discover how AI could achieve this. They conducted experiments in hopes of isolating the specific image features that enabled AI to predict racial identity.

“Overall, we were unable to isolate specific image features that were responsible for the recognition of racial identity in medical images, either by spatial location, in the frequency domain, or that were caused by common anatomic and phenotype confounders associated with racial identity,” the researchers wrote.

The researchers say that AI deep learning models for medical image analysis should be used “with extreme caution as such information could be misused to perpetuate or even worsen the well documented racial disparities that exist in medical practice.”

The groundbreaking research was performed by a global team of scientists affiliated with Stanford University School of Medicine, Massachusetts Institute of Technology (MIT), Indiana University–Purdue University, the University of Toronto, the Vector Institute for Artificial Intelligence in Toronto, Emory University, National Tsing Hua University, Georgia Institute of Technology, University of Adelaide, Arizona State University, Dupage Medical Group, Florida State University College of Medicine, and Beth Israel Deaconess Medical Center in Boston.

Copyright © 2022 Cami Rosso. All rights reserved.