Print  Close

MiDeFace: A FreeSurfer tool for Minimally Invasive DeFacing of MRI and PET data

Poster No:

1909 

Submission Type:

Abstract Submission 

Authors:

Martin Norgaard1,2, Liam Sennott3, Teah Serani3, Nathan Draudt3, Anthony Galassi2, Murat Bilgel4, Cyril Pernet5, Melanie Ganz6,1, Douglas Greve3

Institutions:

1University of Copenhagen, Copenhagen, Denmark, 2NIMH Intramural Research Program, Bethesda, MD, 3Martinos Center for Biomedical Imaging at MGH, Boston, MA, 4NIMH Intramural Research Program, Bethesda, MA, 5Neurobiology Research Unit, Copenhagen, Denmark, 6Rigshospitalet, Copenhagen, Copenhagen

First Author:

Martin Norgaard  
University of Copenhagen|NIMH Intramural Research Program
Copenhagen, Denmark|Bethesda, MD

Co-Author(s):

Liam Sennott  
Martinos Center for Biomedical Imaging at MGH
Boston, MA
Teah Serani  
Martinos Center for Biomedical Imaging at MGH
Boston, MA
Nathan Draudt  
Martinos Center for Biomedical Imaging at MGH
Boston, MA
Anthony Galassi  
NIMH Intramural Research Program
Bethesda, MD
Murat Bilgel  
NIMH Intramural Research Program
Bethesda, MA
Cyril Pernet, PhD  
Neurobiology Research Unit
Copenhagen, Denmark
Melanie Ganz  
Rigshospitalet|University of Copenhagen
Copenhagen, Copenhagen|Copenhagen, Denmark
Douglas Greve  
Martinos Center for Biomedical Imaging at MGH
Boston, MA
E-Poster

Introduction:

MiDeFace (https://surfer.nmr.mgh.harvard.edu/fswiki/MiDeFace) is a novel defacing pipeline developed for the FreeSurfer software suite, aimed at removing identifiable facial features from MRI [2, 3]. This tool addresses the crucial need for anonymizing MRI data by removing identifiable facial features, thus protecting patient privacy without rendering the data unusable for future analyses.The tool can also be extended to deface multi-modal data (such as PET), allowing for specialized applications to be built for specific multi-modal imaging data.

Methods:

The MiDeFace workflow uses the Sequence Adaptive Multimodal Segmentation (SAMSEG) pipeline available in FreeSurfer to segment the entire head, including the skull and eyeballs [1,5]. MiDeFace also uses a surface-based face atlas, where critical facial features (eyes, nose, mouth, cheeks, chin, ears) are labeled and minimally removed in a manner that is hard to reverse (i.e. not able to de-identify). The defacing process includes inward and outward projection of the atlas face, respecting the labeling of the brain and skull.

Results:

The MiDeFace defacing pipeline applies random intensity correction within the bounds of the true data's mean and standard deviation, effectively removing the facial surface while retaining statistical similarity (Figure 1). A study involving 41 subjects demonstrated that MiDeFace's defacing workflow, when compared with unaltered FreeSurfer analyses, showed no significant differences in regions of interest, except a minor reduction (0.1%) in estimated intracranial volume. MiDeFace takes less than 8 minutes to run per subject. Finally, MiDeFace can also be extended to multi-modal data applications, such as PETDeface, a BIDS application for removing facial features in PET data (https://github.com/openneuropet/petdeface) [4].
Supporting Image: figure1_mideface_ohbm2024.png
 

Conclusions:

MiDeFace efficiently anonymizes neuroimaging data, and provides a robust solution for privacy concerns in medical imaging. Its integration into FreeSurfer, along with its capability to work across multiple modalities and create masks applicable to various volumes, makes it a versatile and essential tool for researchers requiring anonymous imaging data while preserving the utility for subsequent neuroimaging analyses.

Modeling and Analysis Methods:

Image Registration and Computational Anatomy
Methods Development 1
Segmentation and Parcellation 2

Novel Imaging Acquisition Methods:

Anatomical MRI
PET

Keywords:

MRI
Open-Source Software
Positron Emission Tomography (PET)
Segmentation
Spatial Normalization
Spatial Warping
STRUCTURAL MRI

1|2Indicates the priority used for review

Provide references using author date format

[1] Cerri, S. (2023), An open-source tool for longitudinal whole-brain and white matter lesion segmentation, NeuroImage: Clinical, Volume 38, 103354, ISSN 2213-1582, https://doi.org/10.1016/j.nicl.2023.103354.

[2] Dale A (1999). Cortical Surface-Based Analysis: I. Segmentation and Surface Reconstruction. Neuroimage. 9(2):179–94. doi:10.1006/nimg.1998.0395.

[3] Fischl B. (2012) FreeSurfer. Neuroimage. 15;62(2):774-81. doi: 10.1016/j.neuroimage.2012.01.021. Epub 2012 Jan 10. PMID: 22248573; PMCID: PMC3685476.

[4] Gorgolewski, Krzysztof J. (2016). Nipype: a flexible, lightweight and extensible neuroimaging data processing framework in Python. 0.12.0-rc1. Zenodo. 10.5281/zenodo.50186

[5] Puonti, O., (2016). NeuroImage Fast and sequence-adaptive whole-brain segmentation using parametric Bayesian modeling. NeuroImage, 143, 235–249. https://doi.org/10.1016/j.neuroimage.2016.09.011