• AI
  • Molecular Imaging
  • CT
  • X-Ray
  • Ultrasound
  • MRI
  • Facility Management
  • Mammography

Reconstructions of radiologic facial images could strip away patient privacy

Article

Do you think stripping out textual identifying information in publicly available radiologic images will protect you against privacy violations? Think again. A paper presented Thursday at the 2010 RSNA meeting showed how facial images reconstructed from maxillofacial sinus and cerebral vasculature images could be matched in a database using commonly available face-matching software.

Do you think stripping out textual identifying information in publicly available radiologic images will protect you against privacy violations? Think again. A paper presented Thursday at the 2010 RSNA meeting showed how facial images reconstructed from maxillofacial sinus and cerebral vasculature images could be matched in a database using commonly available face-matching software. 

The study mixed 29 faces developed using radiology postprocessing software with three sets of facial databases and then used a face-recognition program, Picasa 3.6, to look for matches. The software program hit the mark 27.5% of the time in each of the databases, said Dr. Joseph Mazura of Weill Cornell Medical College.

“For a match to be accurate, two things needed to occur. Picasa must recognize an image as a face and then must match it with the with the subject’s photograph,” Mazura said. In all phases (the databases were expanded 50 images at a time, to include a final total of 179 with the reconstructions), reconstructions were consistently matched or not matched at a 27.5% rate with an upper confidence level of 40.5%.

“Importantly, no 3D reconstruction was matched incorrectly and Picasa did recognize 28 of 29 reconstructions as a face,” Mazura said.

“Software reconstructions of CTs contain sufficient detail to allow matching of subjects’ faces in one out of three cases,” Mazura said. “This suggests that subjects whose head CT data resides in research repositories have the potential to be identified, even if protected textual data is removed. If accuracy rates can be improved significantly, additional de-identification will be required to distort the soft tissue components. Perhaps this is a new way to think about privacy.”

A member of the audience suggested the problem may be overstated. Things such as facial hair, glasses, and the use of profile images can make facial recognition software unusable, she said. If it’s one in three cases with perfect conditions, why should you care, because you rarely have perfect conditions?

Mazura said government agencies are imposing guidelines to make facial image displays more consistent.

“Think about passport or DMV databases,” he said. “I feel there are databases of these photographs that can potentially be used.”

Recent Videos
Current and Emerging Insights on AI in Breast Imaging: An Interview with Mark Traill, Part 1
Addressing Cybersecurity Issues in Radiology
Computed Tomography Study Shows Emergence of Silicosis in Engineered Stone Countertop Workers
Can an Emerging AI Software for DBT Help Reduce Disparities in Breast Cancer Screening?
Skeletal Muscle Loss and Dementia: What Emerging MRI Research Reveals
Magnetoencephalopathy Study Suggests Link Between Concussions and Slower Aperiodic Activity in Adolescent Football Players
Can Diffusion Microstructural Imaging Provide Insights into Long Covid Beyond Conventional MRI?
Assessing the Impact of Radiology Workforce Shortages in Rural Communities
Emerging MRI and PET Research Reveals Link Between Visceral Abdominal Fat and Early Signs of Alzheimer’s Disease
Reimbursement Challenges in Radiology: An Interview with Richard Heller, MD
Related Content
© 2024 MJH Life Sciences

All rights reserved.