Until recently, the "zig zag" curves from the recordings of brain activity by means of electroencephalography (EEG) have been considered unidentifiable. Recent advances in artifical intelligence have demonstrated, that deep neural networks can recognize the recordings from individual patients. This finding puts the anonymity of patients in public datasets at risk.
In this project, we aim to research how EEG recordings can be anonymized and what implications result for publicly sharing EEG data.
The follow-up project, AVATAR-Transfer, addresses the anonymization of complex biosignals, in particular electroencephalograms (EEG), which may contain potentially person-identifying patterns. With the increasing availability of such data in electronic health records, home applications, and cloud services, the risk of re-identification is rising significantly. The aim is to research and validate methods that allow for a variable degree of anonymization while ensuring clinical usability as well as applicability in research and AI-based analysis.
The development of a software toolbox is planned to enable EEG data to be anonymized prior to sharing. In addition, the impact of anonymization on diagnostic assessment by neurologists and on the performance of AI systems will be investigated. Digital watermarks are intended to enable traceability of shared data. Furthermore, the risk of re-identification will be studied by matching clinical EEGs with data from consumer and VR devices.
