Anonymization of Facial Features in MRI Volumes

Thesis by: Christoph Niedermayr
Advisor: NassirNavab
Supervision by: Christian Lederer (TRIUM), AndreasKeil
Due date:

Abstract

The goal of this thesis is to detect and modify facial features in volume MRI scans, allowing hospitals to publish diagnostic imagery without compromising their patients' privacy. The problem can be splitted into three major challenges:

• classify image as head or non-head
• detect facial features
• modify image data where necessary

In order to detect facial features, a given image will be registered against a reference image containing a head. Registration will be done in three phases, increasing the degrees of freedom as the registration improves. In the reference image, crucial areas (e.g. eyes, jaw, nose) are marked and can be mapped to the test image after registration.

Image classification will be done using a statistic model, based on image properties and registration outcome.

In the end, the image data can be manipulated by e.g. deforming the crucial regions.

Literature

• J. B. A. Maintz and M. A. Viergever. An overview of medical image registration methods, 1996.
• W. M. Wells, P. Viola, A. Hideki, N. Shin, and R. Kikinis. Multi-modal volume registration by maximization of mutual information. Medical Image Analysis, 1(1):35-51, 1996.
• D. Mattes, Haynor D. R., H. Vesselle, T. K. Lewellen, and W. Eubank. Nonrigid multimodality image registration. Proceedings of SPIE, 4322:1609-1620, 2001
• L. Ibanez and W. Schroeder. The ITK Software Guide 2.4. Kitware, Inc., 2005