Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2024 Nov;20(11):8048-8061.
doi: 10.1002/alz.14303. Epub 2024 Oct 11.

Implementation and validation of face de-identification (de-facing) in ADNI4

Christopher G Schwarz  1 , Mark Choe  2 , Stephanie Rossi  3 , Sandhitsu R Das  4 , Ranjit Ittyerah  5 , Evan Fletcher  6 , Pauline Maillard  6 , Baljeet Singh  6 , Danielle J Harvey  7 , Ian B Malone  8 , Lloyd Prosser  8 , Matthew L Senjem  9 , Leonard C Matoush  9 , Chadwick P Ward  1 , Carl M Prakaashana  1 , Susan M Landau  10 , Robert A Koeppe  11 , JiaQie Lee  10 , Charles DeCarli  6 , Michael W Weiner  3 , Clifford R Jack Jr  1 , William J Jagust  10 , Paul A Yushkevich  5 , Duygu Tosun  2   3 ; Alzheimer's Disease Neuroimaging Initiative
Affiliations

Implementation and validation of face de-identification (de-facing) in ADNI4

Christopher G Schwarz et al. Alzheimers Dement. 2024 Nov.

Abstract

Introduction: Recent technological advances have increased the risk that de-identified brain images could be re-identified from face imagery. The Alzheimer's Disease Neuroimaging Initiative (ADNI) is a leading source of publicly available de-identified brain imaging, who quickly acted to protect participants' privacy.

Methods: An independent expert committee evaluated 11 face-deidentification ("de-facing") methods and selected four for formal testing.

Results: Effects of de-facing on brain measurements were comparable across methods and sufficiently small to recommend de-facing in ADNI. The committee ultimately recommended mri_reface for advantages in reliability, and for some practical considerations. ADNI leadership approved the committee's recommendation, beginning in ADNI4.

Discussion: ADNI4 de-faces all applicable brain images before subsequent pre-processing, analyses, and public release. Trained analysts inspect de-faced images to confirm complete face removal and complete non-modification of brain. This paper details the history of the algorithm selection process and extensive validation, then describes the production workflows for de-facing in ADNI.

Highlights: ADNI is implementing "de-facing" of MRI and PET beginning in ADNI4. "De-facing" alters face imagery in brain images to help protect privacy. Four algorithms were extensively compared for ADNI and mri_reface was chosen. Validation confirms mri_reface is robust and effective for ADNI sequences. Validation confirms mri_reface negligibly affects ADNI brain measurements.

Keywords: ADNI; anonymization; de‐facing; de‐identification; face recognition.

PubMed Disclaimer

Conflict of interest statement

Christopher G. Schwarz: receives research funding from the NIH, related and unrelated to this work. Mark Choe: reports no disclosures. Stephanie Rossi: reports no disclosures. Sandhitsu R. Das: reports no disclosures. Ranjit Ittyerah: reports no disclosures. Evan Fletcher: reports no disclosures. Pauline Maillard: reports no disclosures. Baljeet Singh: reports no disclosures. Danielle J. Harvey: receives research funding from the NIH, related and unrelated to this work, serves as statistical advisor to PLOS ONE. Ian B. Malone: is supported by grants to his institution from NIH and is an employee of the Dementia Research Centre which is supported by Alzheimer's Research UK, Brain Research Trust, and The Wolfson Foundation. Lloyd Prosser: works for the Dementia Research Centre (DRC) which is supported by: Alzheimer's Research UK; Brain Research Trust; the Wolfson Foundation. L. Prosser reports no additional disclosures to those that support the DRC. Matthew L. Senjem: reports no disclosures. Leonard C. Matoush: reports no disclosures. Chadwick P. Ward: reports no disclosures. Carl M. Prakaashana: reports no disclosures. Susan M. Landau: receives research funding from the NIH, related and unrelated to this work, is on the DSMB/SAB for KeifeRx and the NIH IPAT study, has received speaking honoraria from Eisai and IMPACT‐AD, has consulted for Banner Health, Vaccinex, and has received travel funding and other research support from IMPACT AD and the Alzheimer's Association. Robert A. Koeppe: reports no disclosures. JiaQie Lee: reports no disclosures. Charles DeCarli: reports no disclosures. Michael W. Weiner: serves on Editorial Boards for Alzheimer's & Dementia, and the Journal for Prevention of Alzheimer's Disease (JPAD). He has served on Advisory Boards for Acumen Pharmaceutical, Alzheon, Inc., Amsterdam UMC; MIRIADE, Cerecin, Merck Sharp & Dohme Corp., NC Registry for Brain Health, and REGEnLIFE. He also serves on the USC ACTC grant which receives funding from Eisai. He has provided consulting to Boxer Capital, LLC, Cerecin, Inc., Clario, Dementia Society of Japan, Dolby Family Ventures, Eisai, Guidepoint, Health and Wellness Partners, Indiana University, LCN Consulting, MEDA Corp., Merck Sharp & Dohme Corp., NC Registry for Brain Health, Prova Education, T3D Therapeutics, University of Southern California (USC), and WebMD. He has acted as a speaker/lecturer for China Association for Alzheimer's Disease (CAAD) and Taipei Medical University, as well as a speaker/lecturer with academic travel funding provided by: AD/PD Congress, Amsterdam UMC, Cleveland Clinic, CTAD Congress, Foundation of Learning; Health Society (Japan), Kenes, U. Penn, U. Toulouse, Japan Society for Dementia Research, Korean Dementia Society, Merck Sharp & Dohme Corp., National Center for Geriatrics and Gerontology (NCGG; Japan), University of Southern California (USC). He holds stock options with Alzeca, Alzheon, Inc., ALZPath, Inc., and Anven. Dr Weiner received support for his research from the following funding sources: National Institutes of Health (NIH)/NINDS/National Institute on Aging (NIA), Department of Defense (DOD), California Department of Public Health (CDPH), University of Michigan, Siemens, Biogen, Hillblom Foundation, Alzheimer's Association, Johnson & Johnson, Kevin and Connie Shanahan, GE, VUmc, Australian Catholic University (HBI‐BHR), The Stroke Foundation, and the Veterans Administration. Clifford R. Jack, Jr.: reports no disclosures. William J. Jagust: Jagust receives research funding from the NIH and Genentech, has consulted for Lilly, Biogen, Clario and Eisai and holds equity in Molecular Medicine and Optoceutics. Paul A. Yushkevich: receives research funding from the NIH, related and unrelated to this work. Duygu Tosun: receives research funding from the National Institute on Aging (NIH). Author disclosures are available in the supporting information.

Figures

FIGURE 1
FIGURE 1
Face reconstructions (surface renderings) of example outputs from the four de‐facing programs tested in the preliminary algorithm comparison phase of the analyses. Due to privacy concerns, we do not show reconstructions of unmodified images in this paper.
FIGURE 2
FIGURE 2
Interface developed at University of Pennsylvania for visual evaluation of de‐facing in this work. Face reconstructions from unmodified images were shown on the left, and from de‐faced images in the center, and axial slices from the de‐faced images on the right. Users could freely rotate the face reconstructions.
FIGURE 3
FIGURE 3
Three independent raters scored their subjective assessment of recognizability of six facial features in images de‐faced with each candidate program, using a Likert scale (1 = least recognizable, 5 = most recognizable). Top: Their ratings for each of the facial features individually; Center: Their ratings summed across all six facial features. mri_reface and fsl_deface were favored because the other two did not remove ears. fsl_deface was considered least recognizable for most cases because the results look less like faces at all, versus the other methods, but it occasionally failed and left the face partly intact. mri_reface had the fewest failures at removing the complete face. Bottom: Three independent raters scored their subjective assessment of brain integrity after de‐identification in images de‐faced with each candidate program, using a Likert scale (1 = least preserved; 5 = most preserved). All methods mostly preserved the brain, but mri_reface had the fewest accidental brain modifications, and fsl_deface commonly removed some superior brain voxels.
FIGURE 4
FIGURE 4
Intraclass correlation coefficient (ICC) and bias (subtractive difference %) of measurements from unmodified vs. de‐faced images, from each of the four candidate programs. Top: Left two: effects of de‐facing on subcortical volume measurements; right two: effects on cortical thickness measurements, both from longitudinal FreeSurfer 5.1. . Center: Left two: effects on entorhinal cortex (ERC) volumes from the Automated Segmentation of Hippocampal Subfields (ASHS) pipeline at University of Pennsylvania; right two: effects on longitudinal measurements of atrophy in the parahippocampal cortex from the tensor‐based morphometry with symmetric normalization (TBM‐SyN) pipeline at Mayo Clinic. Bottom: Effects on brain volume (left two) and boundary shift integral (BSI), both as measured by the BSI pipeline at UCL. For all of these plots, “ns” denotes non‐significance at p > 0.05, “*” denotes p ≤ 0.05, “**” denotes p ≤ 0.01, “***” denotes p ≤ 0.001, and “****” denotes p ≤ 0.0001.
FIGURE 5
FIGURE 5
Effects of de‐facing T1‐w and T2‐FLAIR‐w images with mri_reface on the UC Davis WMH pipeline.
FIGURE 6
FIGURE 6
Effects of de‐facing T1‐w MRI and “raw” PET images with mri_reface on PET quantification pipelines at UC Berkeley, after their ADNI standard preprocessing at U. Michigan. FBB, florbetaben; FBP, florbetapir; FTP, flortaucipir; EroSWM, eroded supratentorial white matter; CereWhole, whole cerebellum; CereInf, inferior cerebellum.
FIGURE 7
FIGURE 7
Flowchart of ADNI image, image analysis, and QC data, as it relates to the de‐facing process.
See this image and copyright information in PMC

References

    1. Schwarz CG, Kremers WK, Therneau TM, et al. Identification of anonyvmous MRI research participants with face recognition software. N Engl J Med. 2019. - PMC - PubMed
    1. Provenzano D, Loew MH, Goyal S, Rao YJ. Unmasking a privacy concern: potential identification of patients in an immobilization mask from 3‐dimensional reconstructions of simulation computed tomography. Practical Radiation Oncology. 2021:S1879850021002356. doi:10.1016/j.prro.2021.09.006 - DOI - PubMed
    1. Schwarz CG, Kremers WK, Lowe VJ, et al. Face recognition from research brain PET: an unexpected PET problem. Neuroimage. 2022;258:119357. doi:10.1016/j.neuroimage.2022.119357 - DOI - PMC - PubMed
    1. Schwarz CG, Kremers WK, Arani A, et al. A face‐off of MRI research sequences by their need for de‐facing. Neuroimage. 2023;276:120199. doi:10.1016/j.neuroimage.2023.120199 - DOI - PMC - PubMed
    1. Prior FW, Brunsden B, Hildebolt C, et al. Facial recognition from volume‐rendered magnetic resonance imaging data. IEEE Trans Inf Technol Biomed. 2009;13:5‐9. doi:10.1109/TITB.2008.2003335 - DOI - PubMed

Publication types