Protecting the Confidentiality of Participants in Studies of Alzheimer's Disease and Related Dementias by Replacing Face Imagery in MRI

PROJECT SUMMARY / ABSTRACT There exists a growing demand to share all publicly-funded research data, including magnetic resonance images (MRI). Concurrently, it has been shown that high-resolution facial reconstructions can be generated from MRI, and face recognition software can match these reconstructions with participant photos. Standard MRI de-identification removes participant names from the image header, but does nothing to prevent face recognition. Identified individual research participants would be irreversibly linked with all the collected protected health information, such as diagnoses, biomarker results, genetic risk factors, and neuropsychiatric testing. Although data use agreements can legally protect study administrators, these legal mechanisms do not directly protect participants. If participants were publicly identified by a careless or malicious individual, this event would significantly and permanently erode public trust and participation in medical research. Many large imaging studies of Alzheimer's Disease (AD) and related dementias are vulnerable to this threat. To address this threat, we propose a novel technique that de-identifies MRI by replacing facial imagery with a generic, average face (i.e., a digital face “transplant”). Unlike existing methods that remove or blur faces, our approach minimizes added bias and noise in imaging biomarker measurements by producing a de-identified MRI that resembles a natural image. This imminent privacy threat grows with burgeoning technology and with the increased public sharing of research data. We propose to: improve our de-identification software by collaborating with a top expert in face recognition; further reduce effects on brain measurements; large-scale test/validate on Mayo Clinic aging studies; add capability for de-facing additional imaging modalities; test and improve performance when applied to diverse populations; and share the software freely for research use. Aim 1: Refine and validate an optimized face de-identification algorithm: 1A) Further improve de- identification performance; 1B) Further reduce impacts on brain biomarker measurements; 1C) Test and validate using images from the Mayo Clinic Study of Aging and Alzheimer's Disease Research Center studies. Aim 2: Add capability for de-identifying additional imaging sequences and modalities: 2A) Support additional MRI sequences; 2B) Support PET images; 2C) Support CT images. Aim 3: Investigate effects of age, race, and sex: 3A) Evaluate the effects of age, race, and sex on the proposed de-identification method; 3B) Adapt software to ensure that the algorithm protects all participants equally. Aim 4: Disseminate software and educational materials: 4A) Share the software freely for research use; 4B) Develop and disseminate materials and recommendations for research studies for protection of participant privacy.