Decoding mental states from magnetoencephalographic signals

Andrey Zhdanov
Host Institution: 

Welcome to the joint seminar BioN and VASNei 4.03.2013 at 17.30, Aud 90, Main Building of Saint Petersburg State University (SPbSU)

Decoding mental states from magnetoencephalographic signals.

Andrey Zhdanov,Ph.D. student, Department of Biomedical Engineering and Computational Science,Aalto University

"Brain reading" - inferring personвАЩs mental state from neuropsysiological signals - is an increasingly popular approach to brain research. Historically, it was most extensively pursued in the field of EEG-based brain-computer interface (BCI) design with the earliest works dating back about 40 years [1], [2].
More recently this approach has been increasingly gaining traction in the mainstream functional neuroimaging research, especially in the field of functional Magnetic Resonance Imaging (fMRI) [3]. One particularly interesting possibility is using this technique for investigating bistable perceptual states. Bistable perceptual states (Necker cube and Rubin's vase being among the most prominent examples) are characterized by person's percept spontaneously alternating between two possible interpretations of the same stimulus. Thus they provide a unique opportunity for decoupling physical properties of the stimulus from the subjective experience. Application of the "brain reading" methodology to the investigation of such states has already yielded some interesting results [4].
With very few exceptions all of the "brain reading" studies employ some form of machine learning techniques for extracting the information about the mental state from the brain data. Employing machine learning techniques for this kind of data analysis poses a unique set of challenges such as the demand for robust performance on a noisy high-dimensional data with a limited number of samples and the requirement for interpretability of the resulting models in a physiologically meaningful way.
In my presentation I survey the current progress in application of machine learning techniques in the field of "brain reading" and describe our own experiments in applying regularized Fisher Linear Discriminant Analysis to the problem of inferring the mental state from magnetoencephalographic (MEG) signals. I discuss the motivation for our experiment, the way we address the high dimensionality of data with Tiknonov regularization, and the interpretability of the resulting classifiers [5], [6].
The research described in the presentation was carried out at the Functional Brain Imaging Unit at Tel Aviv Sourasky Medical Center and the School of Computer Science of Tel Aviv University, in collaboration with Prof. Nathan Intrator, Prof. Talma Handler and Prof. Leslie Ungerleider.


[1] Vidal, J. J. Toward Direct Brain-Computer Communication Annual Review of Biophysics and Bioengineering, 1973, 2, 157-180

[2] Vidal, J. J. Real-Time Detection of Brain Events in EEG Proceedings of the IEEE, 1977, 65, 633-641

[3] Haynes, J.-D. & Rees, G. Decoding mental states from brain activity in humans Nature Reviews Neuroscience, 2006, 7, 523-534

[4] Haynes, J.-D. & Rees, G. Predicting the orientation of the invisible stimuli from activity in human primary visual cortex Nature Neuroscience, 2005, 8, 686-691

[5] Zhdanov, A.; Hendler, T.; Ungerleider, L. & Intrator, N. Inferring Functional Brain States Using Temporal Evolution of Regularized Classifiers Computational Intelligence and Neuroscience, 2007, 2007, Article ID 52609, 8 pages

[6] Zhdanov, A.; Hendler, T.; Ungerleider, L. & Intrator, N. Machine Learning Framework for Inferring Cognitive State From Magnetoencephalographic (MEG) Signals Proceedings of the 1st International Conference on Cognitive Neurodynamics (ICCN'07), 2007, 393-397