Related Experiment Video
Updated: Jul 29, 2025
08:15
Reverse Dissection and DiceCT Reveal Otherwise Hidden Data in the Evolution of the Primate Face
Published on: January 7, 2019
7.0K
De-Identification Technique with Facial Deformation in Head CT Images.
Tatsuya Uchida1, Taichi Kin2,3, Toki Saito4
1Department of Neurosurgery, The University of Tokyo, 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8655, Japan.
Neuroinformatics
|May 24, 2023
Summary
A new de-identification technique distorts faces in head CT scans, preventing identification while preserving crucial medical data. This method ensures patient privacy without compromising diagnostic accuracy for AI models.
Area of Science:
- Medical Imaging
- Computer Vision
- Data Privacy
Background:
- Head CT scans, including facial regions, allow 3D reconstruction, posing risks of patient identification.
- Existing de-identification methods may compromise image data integrity.
Purpose of the Study:
- To develop and evaluate a novel de-identification technique for head CT images.
- To ensure facial distortion does not impact the accuracy of deep learning models for medical analysis.
Main Methods:
- A de-identification technique was developed by deforming facial regions of head CT images using control points.
- Face detection and identification programs assessed identification rates and confidence scores.
- Intracranial volume equivalence tests and pixel value histogram correlations were performed.
- Deep learning model accuracy for intracranial segmentation was evaluated using Dice Similarity Coefficient.
Main Results:
- The de-identification technique achieved a 100% face detection rate with confidence scores below 90%.
- Statistical equivalence was confirmed for intracranial volume before and after deformation.
- High similarity (median correlation coefficient 0.9965) was observed in intracranial pixel value histograms.
- Dice Similarity Coefficient values remained statistically equivalent, indicating preserved accuracy.
Conclusions:
- The developed technique effectively de-identifies head CT images by distorting facial features.
- This method successfully prevents facial identification while maintaining the integrity of intracranial information for deep learning applications.
