1. |
- Abramian, David, 1992-, et al.
(author)
-
REFACING: RECONSTRUCTING ANONYMIZED FACIAL FEATURES USING GANS
- 2019
-
In: 2019 IEEE 16TH INTERNATIONAL SYMPOSIUM ON BIOMEDICAL IMAGING (ISBI 2019). - : IEEE. - 9781538636411 ; , s. 1104-1108
-
Conference paper (peer-reviewed)abstract
- Anonymization of medical images is necessary for protecting the identity of the test subjects, and is therefore an essential step in data sharing. However, recent developments in deep learning may raise the bar on the amount of distortion that needs to be applied to guarantee anonymity. To test such possibilities, we have applied the novel CycleGAN unsupervised image-to-image translation framework on sagittal slices of T1 MR images, in order to reconstruct, facial features from anonymized data. We applied the CycleGAN framework on both face-blurred and face-removed images. Our results show that face blurring may not provide adequate protection against malicious attempts at identifying the subjects, while face removal provides more robust anonymization, but is still partially reversible.
|
|
2. |
- Tarun, Anjali, et al.
(author)
-
Graph spectral analysis of voxel-wise brain graphs from diffusion-weighted mri
- 2019
-
In: 2019 IEEE 16th International Symposium on Biomedical Imaging (ISBI 2019). - : IEEE. - 9781538636411 ; , s. 159-163
-
Conference paper (peer-reviewed)abstract
- Non-invasive characterization of brain structure has been made possible by the introduction of magnetic resonance imaging (MRI). Graph modeling of structural connectivity has been useful, but is often limited to defining nodes as regions from a brain atlas. Here, we propose two methods for encoding structural connectivity in a huge brain graph at the voxel-level resolution (i.e., 850'000 voxels) based on diffusion tensor imaging (DTI) and the orientation density functions (ODF), respectively. The eigendecomposition of the brain graph's Laplacian operator is then showing highly-resolved eigenmodes that reflect distributed structural features which are in good correspondence with major white matter tracks. To investigate the intrinsic dimensionality of eigenspace across subjects, we used a Procrustes validation that characterizes inter-subject variability. We found that the ODF approach using 3-neighborhood captures the most in-formation from the diffusion-weighted MRI. The proposed methods open a wide range of possibilities for new research avenues, especially in the field of graph signal processing applied to functional brain imaging.
|
|