Data that’s supposedly anonymous has a nasty habit of becoming identifiable again. Institutions believe that we can ‘anonymize’ data by taking key bits of information out, and ‘pseudoanonymization’ is even a recommendation in the GDOR, but that doesn’t always protect peoples’ identities. We’ve seen researchers using entropy to identify people in these anonymized documents for well over a decade.
Now, a study from the Mayo Clinic has shown that anonymizing medical image files may not be enough to render them private either.
Magnetic resonance imaging (MRI) scans and other medical images include data such as the patient’s name, the date of the scan, and the identification number. Healthcare facilities delete this information before sharing the data with other organizations, but researchers have been able to deduce the identities of patients by manipulating the facial images that show up on the scans.
The researchers, who published their results as a letter in the New England Journal of Medicine, recruited 84 volunteers between 34 and 89 years old who had undergone a head MRI within the prior three months.
They photographed each participant’s face from five varying angles. Then, they used each MRI to reconstruct the patient’s face and make ten 2D images with random lighting. They used the facial recognition algorithms that Microsoft offers via its Azure cloud service to try and match the reconstructed images against the original photographs. The correct MRI scan was the number one match in 70 of the 84 cases (83%).
“The current standard of removing only meta-data in medical images may be insufficient to prevent reidentification of participants in research,” the researchers said in their letter, adding: “This identification would result in an infringement of privacy that could include diagnoses, cognitive scores, genetic data, bio-markers, results of other imaging, and participation in studies or trials.”
You could argue that MRI images are under tight lock and key, so using this technique in anger would be a stretch. In reality, though, that isn’t the case. A ProPublica study conducted in September this year found MRIs, X-rays and CT scans belonging to around five million Americans and millions more around the world sitting unprotected on the Internet. Much like biometric information, once that data is compromised, there’s no taking it back.
This illustrates the need for foresight in data collection and sharing. Even the most innocuous of documents, supposedly scrubbed clean, could still leave a lingering trace of a person’s identity for someone with the resources and commitment to find it.
