Facial Recognition Software Identifies Patients From MRI Scans

Reaction to date suggest there are more concerns about the implications of privacy threats than any medical breakthrough.

Advances in technological power assisted by AI increasingly render anonymity impossible to maintain. JL

Melanie Evans reports in the Wall Street Journal:

Facial-recognition software correctly matched photos of research volunteers to unidentified medical scans of their heads, in a new study of images that are commonly used in brain research. The finding draws attention to a privacy threat that will increase with technology improvements and the growth of health-care data. Information on the identities of participants is typically kept separate from the data, to protect patients’ privacy as databases of information get shared. But other studies have identified people from genetic information and data from wearable activity-monitoring devices that had been stripped of personal information.
Facial-recognition software correctly matched photos of research volunteers to unidentified medical scans of their heads, in a new study of images that are commonly used in brain research.
The finding draws attention to a privacy threat that will increase with technology improvements and the growth of health-care data, experts in medical imaging and facial recognition said.

Researchers at the Mayo Clinic used commercially available facial-recognition software to match photographs of about 80 volunteers to unidentified MRI images that show outlines of the head in addition to the brain itself. The software correctly matched 83% of the images, they reported in the New England Journal of Medicine.
Results are the latest to find technology has outflanked privacy protections in health care, where an aggressive push is under way to amass and mine medical data from patient medical records, research, medical devices and consumer technology such as smartwatches.
Information on the identities of participants is typically kept separate from the data, to protect patients’ privacy as databases of information get shared. But other studies have identified people from genetic information and data from wearable activity-monitoring devices that had been stripped of personal information.
The findings suggest existing privacy protections don’t go far enough, the study’s authors said. They also point to the continuing challenge of anticipating new risks from emerging technology, as well as to the need for caution in handling medical data, said I. Glenn Cohen, director of Harvard University’s Petrie-Flom Center for Health Law Policy, Biotechnology and Bioethics. “Our imaginations are only so good,” he said.
Mayo Clinic researchers launched the study after noticing the clarity of images used to look at brain structures in studies on aging, Alzheimer’s disease and dementia, said Christopher Schwarz, a Mayo Clinic computer scientist and researcher, who works with the images and is lead author of the study. “If that was my dad, could I recognize him?” Mayo Clinic researchers speculated, Mr. Schwarz recalled.
The answer was likely yes, he said. That raised concerns about privacy should algorithms be able to pick out faces on a much greater scale. “How bad of a problem is this?” the researchers wondered.
Mayo Clinic recruited 84 volunteers, ages 34 to 89, who had recently received an MRI, or magnetic resonance imaging, scan of the head during a clinic study. Volunteers were photographed from five angles. Researchers reconstructed an image of each face from the MRIs, which capture the outline from skin, fat and bone marrow in the skull but don’t capture bone or hair.
The study then tested facial-recognition software from Microsoft MSFT 1.97% Azure to match an MRI-generated face reconstruction to its owner’s photo. The algorithm correctly matched 70 of 84 images, the study said. For another 10 images, the software ranked the correct photo in its top five candidates.
Microsoft declined to comment.
Results didn’t totally surprise researchers, who themselves saw distinct facial features in MRI images, Mr. Schwarz said. Nonetheless, findings were sobering. “We found that this really is a problem,” he said.
Identifying an MRI used in research could expose private information used in studies, such as family medical history, illnesses, genetic data and risks for diseases, he said. Research studies typically collect personal information from patients and their medical records.
Mayo Clinic researchers hope to develop a fix to publish broadly, Mr. Schwarz said. Alternatives on the market now, such as blurring or removing some data from MRIs, can limit brain-image quality, he said.
Privacy threats from MRIs are likely limited, however, and the study’s test of facial-recognition software avoided some harder challenges found in everyday life, Mr. Schwarz and other researchers who weren’t involved in the study said.
“The risk to the typical patient is really small,” said Eliot Siegel, a radiology professor at the University of Maryland School of Medicine, who has studied the issue but wasn’t involved in the recent research.
Researchers typically pledge not to try to identify medical study participants under data-sharing contracts, he said. Facial-recognition software also likely wouldn’t work as well when searching for matches across thousands of random photos instead of dozens of images.
Still, facial-recognition software continues to make significant advancements in accuracy after a major leap from deep learning in 2014, experts in the technology said. Privacy risks are compounded by rapidly growing amounts of medical data for valuable research, experts in medical and facial-recognition technology said.
“As time goes on, that risk is going to increase and it’s really important to consider that as we continue to develop larger and larger machine-learning databases,” Dr. Siegel said.