RSNA 2012 - Ensuring Research Subject Privacy: Anonymization andObscuring of Facial Features of Shared Head Volumes Using Open SourceOnline Tools in XNAT (LL-INE1243)

One of the tricky things about anonymization of image data for scientific research is the fact that although we de-identify the DICOM header and textual content of the images, an additional issue is involved in CT and MR of the head. This issue involves the fact that based on the data acquired a three dimensional visualization of the face can be reconstructed on which the patient could be identified. This work provides a solution to deface MR data in order to tackle this problem.

Ensuring Research Subject Privacy: Anonymization and Obscuring of Facial Features of Shared Head Volumes Using Open Source Online Tools in XNAT
Mikhail Milchenko, PhD , Kevin Archie , Daniel Marcus, PhD
PURPOSE/AIM
1. To understand anonymization requirements for DICOM tags;
2. To understand whether a 3D rendering of head anatomy scan (e.g. MR) can identify a person;
3. To understand the difference between de-identified and original head scans for the identification purposes, and for the purposes of automatic post-processing.
4. To learn to share MR head data using automatic online de-identification tools based on pipelines in XNAT (open source research image sharing database);
CONTENT ORGANIZATION
1. 3D head visualizations based on high resolution MR scan vs. photographs for identification purposes.
2. Obscuring facial features: registering the image into atlas space combined with anatomical surface obscuring
3. Sharing data using the open source research database engine (XNAT) with DICOM anonymization and face obscuring pipeline
4. Comparison of original and face-obscured MR head renderings.
5. Outcomes of post-processing tools (gain field correction, skull stripping, registration) on original and face-obscured MR images
SUMMARY
The exhibit will review the de-identification technique that consists of DICOM anonymization and obscuring of face anatomy. The emphasis will be made on learning the impact of face obscuring algorithm on visual change in a 3D rendering appearance, and of examining the outcomes of post-processing tools used on obscured images.