In the area of Brain Computer Interfaces, there is plenty of hype and abundant wishful thinking. I have contributed to the froth myself. << link>>
But I read this week about an application of brainscans that is cool and actually makes sense. Maria V. Ruiz-Blondet and colleagues at SUNY Binghamton have reported on ‘CEREBRE’ a system that uses measurements of brain activity as a biometric identification. What catches the eye is the reported accuracy of 100%! 
One key to this result is that instead of a generic EEG, they use stimulus-locked and averaged evoked respoonse potential (ERP). Effectively, they measure the brain’s specific response to visual stimuli, averaging away all the other activity in the EEG. Identification is achieved by measuring response to a set of these stimuli, which are an accurate biometric identifier. I.e., the recordings are unique for each brain.
The identification classified sets of “time locked” EEG measures for different kinds of stimuli. Note that this does not involve feature extraction from the signals, which makes is simpler and faster than feature analysis. These classifiers can also be combined to form a voting ensemble.
The idea for the biometric identification system is to collect reference data, create the classifiers to create the database of users. The identification system would present a subset of the reference stimuli as the challenge, and measure the brainwaves to “classify” the responses as belonging to a given user.
The use of ERP seems to work much better than EEG per se. I would say that the EEG measures too much, and is too sensitive to work as well as the focused and constrained ERP for this purpose.
The authors comment that this approach has some useful features, aside from being hard to fake. The measured response is involuntary, and requires a live, conscious, and healthy individual. The latter is good for personnel—there is no advantage to be gained by an adversary from harming the user. The use of multiple stimuli means that there can be many different challenges, and the identification can be cancelled if necessary (e.g., by changing the challenge stimuli).
Samuel K. Moore reports for IEEE Spectrum that these results were achieved with “high-quality research-grade EEG”. He quotes researcher Laszlo to say that using commercial grade “game controllers”, such as I have used, “the results weren’t spectacular”. (I’m not surprised.)
Pretty cool stuff!
- Ruiz-Blondet, M. V., Z. Jin, and S. Laszlo, CEREBRE: A Novel Method for Very High Accuracy Event-Related Potential Biometric Identification. IEEE Transactions on Information Forensics and Security, 11 (7):1618-1629, 2016.