Facial Recognition From Volume-Rendered Magnetic Resonance Imaging Data

Abstract

Three-dimensional (3-D) reconstructions of computed tomography (CT) and magnetic resonance (MR) brain imaging studies are a routine component of both clinical practice and clinical and translational research. A side effect of such reconstructions is the creation of a potentially recognizable face. The Health Insurance Portability and Accountability Act of 1996 (HIPAA) Privacy Rule requires that individually identifiable health information may not be used for research unless identifiers that may be associated with the health information including "Full face photographic images and other comparable images ..." are removed (de-identification). Thus, a key question is: Are reconstructed facial images comparable to full-face photographs for the purpose of identification? To address this question, MR images were selected from existing research repositories and subjects were asked to pair an MR reconstruction with one of 40 photographs. The chance probability that an observer could match a photograph with its 3-D MR image was 1 in 40 (0.025), and we considered 4 successes out of 40 (4/40, 0.1) to indicate that a subject could identify persons' faces from their 3-D MR images. Forty percent of the subjects were able to successfully match photographs with MR images with success rates higher than the null hypothesis success rate. The Blyth-Still-Casella 95% confidence interval for the 40% success rate was 29%-52%, and the 40% success rate was significantly higher ( P < 0.001) than our null hypothesis success rate of 1 in 10 (0.10).

Obscuring Surface Anatomy in Volumetric Imaging Data

“…Previous research (Prior et al 2009; Chen et al 2007; Budin et al 2008) on our ability to recognize MR or CT datasets as belonging to a specific individual based on head surface renderings indicates that such renderings are poor substitutes for photographs, although chances for such recognition are higher than random. It is hard to model conditions for potential person identification under which MR or CT images are shared for analysis; forced choice recognition evaluation used in studies represents somewhat simpler recognition problem, whereas in real situation one may have to compare a surface rendering with images from memory.…”

“…MR and CT images of the head may be used to create high resolution representations of the face, potentially allowing to identify the human subjects involved in such research. Several studies (Prior et al 2009; Chen et al 2007; Budin et al 2008) examined the ability of human observers to identify a facial anatomy rendering with an image as presented in a photograph set. Budin et al (2008) indicated that the chance of correctly connecting a 3D face rendering with a photographic portrait is higher than random guess, and Prior et al (2009) also reported statistical significance of such probability.…”

“…Several studies (Prior et al 2009; Chen et al 2007; Budin et al 2008) examined the ability of human observers to identify a facial anatomy rendering with an image as presented in a photograph set. Budin et al (2008) indicated that the chance of correctly connecting a 3D face rendering with a photographic portrait is higher than random guess, and Prior et al (2009) also reported statistical significance of such probability. On the other hand, named studies report high difficulty of recognition experienced by raters, which may indicate that, while maintaining a potential privacy risk, facial anatomy renderings are still very poor substitutes for a photograph.…”

Image De-Identification Methods for Clinical Research in the XDS Environment

“…Furthermore, a three dimensional image reconstruction, mostly from CT or MR modalities of the head, combined with facial recognition software may also reveal the identity of patients [ 16 , 17 ]. Several other methods are being investigated to be developed and provide a better protection to the patient data [ 18 ]…”

DeID – a data sharing tool for neuroimaging studies

“…Structural images of the brain typically include voxels representing the face that can be rendered to visualize the face. There is evidence that rendered faces can be identified at above chance rates when subjects were given pictures to match with the rendered faces (Prior et al, 2009 ). Thus, voxels representing the face should be removed to reduce the likelihood of re-identification.…”