Redacting MRI Volumes

Introduction

High resolution structural MRI volumes can potentially be used to identify the individual to which the image is of through facial recognition techniques. With the rise of machine learning techniques and the plethora of images on social media, the automated identification of individuals is now practical. Any data that is being shared or stored/processed on platforms shared with others should have as many identifying features removed as is feasible without preventing it's use in the intended processing pipelines.

Face/ear removal

The minimum recommended de-identification step is the removal of the facial features and the ear canals. If the skull volume is not required for the processing pipeline, then skull stripping should be carried out to achieve this; with FSL you would use BET to achieve this.

Where the skull volume is required then you should take steps to de-face the volume. FSL 6.0.3+ include the tool fsl_deface which can be used for this purpose. A summary of a range of tools and their effectiveness can be found in this paper https://www.sciencedirect.com/science/article/pii/S1053811921001221.