Facial recognition matches images reconstructed from MRIs with 83 percent accuracy
Researchers at the Mayo Clinic have found that they can identify people from MRI images used in brain research and diagnosis with facial recognition software, achieving an 83 percent match rate in tests of about 80 volunteers, The Wall Street Journal reports.
The research results have now been published in the New England Journal of Medicine, and may indicate that existing privacy protections do not go far enough, according to the study authors. Researchers noticed the clarity of images used in another study, and wondered if they could be used to identify individuals.
Volunteers in the study were between 34 and 89 years of age, and were photographed at five angles using magnetic resonance imaging (MRI). A facial image was reconstructed from the MRIs, which do not capture bone or hair, but do outline other kinds of tissue.
Facial recognition technology from Microsoft was used by the researchers to successfully match 70 of the 84 images. The correct photo was also included in the top five candidates for 10 of the 14 others.
Mayo Clinic Computer Scientist and Researcher Christopher Schwartz said the risk to patient privacy “really is a problem.”
There are some technologies on the market to deal with the problem, such as by blurring or removing some data, but those can effect the quality of the image, according to Schwartz.
The current risk to typical patients is very small, University of Maryland School of Medicine Radiology Professor Eliot Siegel says, particularly with larger data sets, but will increase over time as the technology is refined.
Earlier this year Amazon announced the development of technology for removing identifying information from medical images, as advancing technology intended to improve medical care creates new potential vulnerabilities.