Assessment of Defacing Techniques on Medical Images for Radiation Therapy: Implications for Patient Privacy and Data Utility.
Authors
Affiliations (3)
Affiliations (3)
- Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, 19104. Electronic address: [email protected].
- Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, 19104. Electronic address: [email protected].
- Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, 19104.
Abstract
To assess the impact of defacing-based deidentification techniques on reidentification risk and data utility across multimodal imaging in radiotherapy. We applied four defacing techniques: biometric_mask, quickshear, mri_reface, and Carina's deidentifier, to imaging from 88 brain patients (MRI, CT, and RTDose) and 97 head-and-neck (HN) patients (PET, CT, and RTDose) in The Cancer Imaging Archive (TCIA). Reidentification risk was assessed using ArcFace, a deep learning-based facial recognition model, by measuring cosine similarity scores and conducting receiver operating characteristic (ROC) analysis to distinguish between original and defaced images. Data integrity was evaluated by statistically comparing volume and image intensity changes between original and defaced images across nine critical organs and GTV. Quickshear provides the highest privacy protection, achieving the lowest AUC across imaging modalities (AUC: 0.61-0.74), followed by Carina (0.59-0.80). Mri_reface showed moderate protection (0.70-0.91), while biometric_mask offered the least (0.76-0.94). Carina preserved structure volumes, and mri_reface produced minor volumetric changes in the eyes (5%) and lens (9%). In contrast, biometric_mask substantially affected the mandible (39%) and oral cavity (69%), while quickshear significantly altered multiple structures (10.0%-86.6%). Median changes in mean CT intensity after defacing were -48.3% (IQR: 32%) with biometric_mask and -77.7% (IQR: 37.7%) with quickshear in the oral cavity. For eyes, Carina and mri_reface produced changes of +24.5% (IQR: 22.6%) and +54.9% (IQR: 66.1%), respectively. In the brain dataset, biometric_mask and quickshear decreased oral cavity D<sub>mean</sub> by 0.72 Gy (IQR: 1.22) and 1.61 Gy (IQR: 2.84), respectively. Carina and mri_reface reduced Eyes D<sub>max</sub> by 2.11 Gy (IQR: 3.44) and 1.05 Gy (IQR: 1.65), respectively. A similar trend was observed in the HN dataset with larger deviations. Carina's deidentifier and mri_reface showed favorable privacy-utility trade-offs relative to facial removal; the optimal choice may vary by application priorities.