Differentially Private EEG Feature Anonymization: A Privacy-Utility Case Study in Clinical Neurophysiology
Adding noise to brain scans to hide patient identity without ruining diagnosis
When hospitals share EEG brain recordings for research, simply removing names isn't enough to protect patients—sophisticated analysis can still re-identify them. This study tested whether adding carefully calibrated random noise to EEG features can hide individual identity while keeping the data useful for AI diagnosis tools, finding that the approach works but requires precise tuning to avoid destroying the signal's clinical value.
Healthcare systems need to share EEG data to improve AI diagnostic tools, but current privacy safeguards are inadequate for high-dimensional brain recordings. This research provides a practical framework hospitals can use to anonymize EEG data before sharing it with researchers, reducing the risk of patient re-identification while maintaining the data quality needed for developing better clinical decision-support systems.