Mbirn: Proof that defacing algorithm removes subjects identity

• Goals
  • To share human data in compliance with federal, state and local regulations including the recently enacted Health Insurance Portability and Accountability Act of 1996 (HIPAA), it is crucial to have in place robust practices and procedures that protect the welfare of the individuals who participate in the research. These practices must include measures that ensure the privacy of the individual. One of the HIPAA-defined 18 identifiers that must be removed from any data in order for it to qualify as sharable under the “safe harbor” regulations is “Full face photographic images and any comparable images.” With the increasing resolution of morphometric scans, it has become increasingly possible to reconstruct detailed images showing facial anatomy. Thus, automated techniques to obscure or remove an individual’s facial features from structural images have become an important part of the data sharing process for large-scale, multi-site projects such as the Biomedical Informatics Research Network (BIRN). In addition to rendering a subject unidentifiable, a method must be robust, removing only nonbrain tissue, yet leaving the brain tissue intact. Further, it must have no debilitating effects on later data processing and analysis.

• Methods
  • The purpose of the behavioral experiment is to demonstrate that the defacing algorithm is effective in removing identifying features, such that individuals regarding the defaced MR structural image of a familiar or unfamiliar person will be unable to identify the person from this image. Face recognition studies have shown that internal facial features (eyes, nose, mouth) are more important than external features (hair, ears, face outline) for recognizing familiar subjects, whereas both feature classes are of equal importance for identifying unfamiliar subjects (Clutterbuck & Johnston 2002). This is particularly important if the subject is participating not as a normal control, but as a representative diagnosed with a particular illness (e.g., HIV, depression). The behavioral experiment will entail the use of two subject groups; the FAMILIAR subject group, containing persons who know the individuals that have consented to supply data (in the form of a photograph and defaced MR structural image); and the UNFAMILIAR subject group, containing persons who have no personal knowledge or previous interaction with the individuals that have consented to supply data.
  • The datasets to be used for the experiment will include defaced MR structural images and photographs of individuals for a subsequent match paradigm. The subject will be asked to match defaced 3-D renderings of MR images to photographs of individuals and to rate the familiarity of the face depicted in the photograph.

• Data
  • MR/Photo Datasets. Twenty-two photographs and twenty-one MR datasets are available. The MR datasets have been defaced; most have passed visual inspection. Four of the remaining datasets omitted the lower portion of the face in the unaltered version and will be replaced with volumes which do include the whole face.
  • Paradigm. The experimental paradigm to generate faces and MR images for the task will be written in Presentation (http://neurobehavioralsystems.com/). The paradigm development is currently in progress.

• Analyses
  • Results from the behavioral paradigm will be analyzed to determine ability of individuals to match photographs and MR datasets, and will further differentiate results based upon familiarity scores.

---