Home - Emerging Tech - Algorithm Removes Facial Features From Neurological CT Images
Research Brief

Algorithm Removes Facial Features From Neurological CT Images

The approach may help protect patient privacy while improving accessibility to neurological images for collaborative research and education.

By
Jill Sakai, Contributor
Thursday, September 27, 2018

SHARE: 

Maintaining privacy of patient data is a major issue in collaborative research and educational efforts that rely on medical images. This is especially a concern with neurological images, such as computed tomography or magnetic resonance imaging scans of the head and brain, since 3D imaging tools could reconstruct an identifiable image of an individual’s face even if all patient data has been removed.

A presentation at the 2018 IEEE 31st International Symposium on Computer-Based Medical Systems in Sweden in June described a service that can be integrated with standard medical image archiving and communications systems to de-identify patient faces in neurological CT images during the file storage process.

The facial de-identification mechanism applies a filtering algorithm that identifies the portion of each CT image corresponding to brain tissue, then defines a line dividing the face from the rest of the image. The facial portion of the image can then be removed before the remaining information is archived in a standard image database. The “defacing” process substantially increased the amount of time required for image storage, taking 140 seconds per image compared to 4.36 seconds for default storage alone, which the authors note is a one-time operation.

The authors validated the system using a publicly accessible data set of head CT images from 48 patients. Manual inspection of each image post-processing concluded that the approach removed identifying facial features without compromising the quality of the brain image. As image archives move from passive repositories toward sources for active data mining, this approach may improve access to existing data while protecting patient privacy.