University of Minnesota
University Relations
myU OneStop


Biomedical Functional Imaging and Neuroengineering Laboratory

College of Science & Engineering > Dept. of Biomedical Engineering

Neural Sensing and Interfacing

Spatial-Time-Frequency Approach to Neural Sensing & Interfacing

EEG is an important technique for studying the temporal dynamics of neural activities and interactions. The state-of-the-art EEG mapping includes a high-density array of sensors that record electrical potentials over the scalp, giving rise to a spatiotemporal dataset. Our lab has developed a variety of sensing and signal processing techniques to extract the signature of brain activity spanned in time, frequency and spatial domains ( Wang & He, J Neural Eng., 2004, Abstract; Wang et al., Clin. Neurophys., 2004, Abstract ; Qin et al., J Neural Eng, 2005, Abstract ). Statistical approaches such as principal component analysis (PCA) and independent component analysis ( ICA ) are used for denoising and isolating signal sources from the raw EEG data. Spatial filtering techniques such as Laplacian filtering allow the enhancement of spatial resolution of EEG mapping. Time-frequency analysis using wavelet decomposition has been demonstrated to be capable of characterizing event-related (de)synchronization (ERD/ERS) associated with motor i magery tasks. We have been investigating the exploitation of these ERD/ERS signals for high accuracy pattern recognition and classification in the brain-computer interface (Yamawaki et al., IEEE-TNSRE, 2006, pdf) and other neuroscience research. The high efficiency could be achieved by such novel spatial-time-frequency approach through optimizing the information extraction from multi-channel EEGs.

Fig. 1 The figure shows the weights of time-frequency pairs in 4 subjects that contribute to the classification of motor imagery tasks. Note that there are certain variations in this group of subjects over both time and frequency domains. These weight maps represent subject-specific signatures guiding designing of adaptive filters and classifiers for different subjects. (from Wang T, Deng J, He B: "Classifying EEG-based Motor Imagery Tasks by means of Time-frequency Synthesized Spatial Patterns," Clinical Neurophysiology, 115(12): 2744-2753, 2004, with permission from Elsevier)

Imaging Human's "Intention"

We have investigated Brain-Computer Interface (BCI) by means of source imaging. BCI is a method of communication based on voluntary neural activity generated by the brain and independent of its normal output pathways of peripheral nerves and muscles. The neural activity used in BCI can be recorded using invasive or noninvasive techniques. EEG signals, because of their relatively short time constants, are widely used in BCI systems. BCI can provide the brain with a new, non-muscular communication and control channel for conveying messages and commands to the external world. We have been investigating new means of extracting useful information using from scalp recorded single trials of EEG signals corresponding to motor imagery tasks through neural source imaging techniques, which offers high sensitivity of detecting the intent of human subject (Qin et al, J Neural Eng, 2004, Abstract ; Kamousi & He, IEEE-TNSRE, 2005, pdf ; Kamousi et al., J Neural Eng, 2007, Abstract; Yuan et al., IEEE-TNSRE, 2008, Abstract). Such investigation enables us to image the "intention" of human subjects who are performing motor imagery tasks in order to control external device such as a computer cursor. We have also developed a efficient BCI from EEG based on goals. This goal selection BCI promises to increase further the information transfer rate which may have important applications to the practical use of BCI systems (Royer & He, J Neural Eng, 2009, Abstract). A video story with regard to the BCI experiment can also be seen here ( online video ).

Fig. 1 Schematic diagram of a BCI system (from He, ed. Neural Engineering, 2005).

Fig. 2 Cortical distribution of the correlation between the mu rhythm and movement imaginations in four subjects. (from Yuan et al., IEEE-TNSRE, 2008)