A generalisable head MRI defacing pipeline: Evaluation on 2,566 meningioma scans

TL;DR

The paper tackles privacy-preserving access to MRI data by addressing the limitations of existing defacing tools across heterogeneous clinical scans. It introduces a generalisable defacing pipeline that fuses atlas-based registration with brain masking to remove facial features while preserving brain anatomy in native space. On a large, multi-center dataset of $2{,}566$ scans from $105$ patients (total $4{,}932$ MRIs) collected between $2012$ and $2020$, the pipeline achieves a defacing success rate of $99.92\%$, with a brain-mask similarity of $0.9975$ and near-unity label-propagation DSCs across regions ($0.9997$ to $0.9999$). These results demonstrate effective brain-tissue preservation and accurate anatomical labeling, enabling secure, multi-centre sharing of neuroimaging data, with the code openly available at GitHub.

Abstract

Reliable MRI defacing techniques to safeguard patient privacy while preserving brain anatomy are critical for research collaboration. Existing methods often struggle with incomplete defacing or degradation of brain tissue regions. We present a robust, generalisable defacing pipeline for high-resolution MRI that integrates atlas-based registration with brain masking. Our method was evaluated on 2,566 heterogeneous clinical scans for meningioma and achieved a 99.92 per cent success rate (2,564/2,566) upon visual inspection. Excellent anatomical preservation is demonstrated with a Dice similarity coefficient of 0.9975 plus or minus 0.0023 between brain masks automatically extracted from the original and defaced volumes. Source code is available at https://github.com/cai4cai/defacing_pipeline.

Figures (2)

• Figure 1: Comparison of (i) defacing failures with affine registration and (ii) corrected cases using the proposed pipeline.
• Figure 2: Overview of the proposed MRI defacing pipeline. Steps are shown in blue; the rhomboid illustrates voxel orientation before and after reorientation to MNI152 space.