- Email: [email protected]
Spatial gradient of P300 in the brain–computer-interface paradigm
Symposium abstracts / International Journal of Psychophysiology 69 (2008) 139–205
Spatial gradient of P300 in the brain–computer-interface paradigm I.A. Shevelev a, E.S. Mikhailova a, V.A. Chicherov a, V.A. Konishev b, D.V. Karlovskiy b a Institute of Higher Nervous Activity and Neurophysiology, Sensory Physiology Department, Moscow, Russia b Neurobotics Co, Zelenograd, Russia An analysis of the 2-D spatial gradient of P300 square that has an influence on the Brain–Computer-Interface (BCI) effectivity was performed in 16 healthy adult subjects. In the first experiment we studied recognition of different intended Cyrillic letters and words by P300 wave in the VEP. The set of optimal characteristics of visual stimulation by the letter matrix that provided maximal recognition level, as well as the best electrode position (Pz) were determined. It was found that the most effective criteria for ВСI decision were: P300 square, P300 amplitude and covariation coefficient. In the second experiment we measured the square of P300 wave for different location of the intended letters of the matrix. The horizontal and vertical spatial gradients of this square as a function of the angular distance between intended and the other letters and a tuning acuteness of visual attention to a significant letter were measured. We revealed the high acuteness of this tuning (width of the tuning curve at its half height was equal to 1.6grad.) independent of the letter position on the letter matrix. The horizontal and vertical gradients of P300 were found to be very similar, but in the half of cases the first one revealed some kind of the “lateral inhibition”: decrease of P300 square for the columns adjacent to the meaningful one. The tuning acuteness was found to be reliable and directly interrelated with P300 square. The data are evidenced high selectivity of local visual attention to letters and suggested some peculiarities of visual spatial attention to columns and rows of letters matrix.
doi:10.1016/j.ijpsycho.2008.05.477 Efficiency of three algorithms for classifying EEG patterns related to different mental tasks V.Y. Roschin a, M.S. Panko b, O.V. Ponfilenok b, A.V. Obukhov a Institute of Higher Nervous Activity and Neurophysiology RAS, Laboratory of Mathematical Neurobiology of Learning, Moscow, Russia b Moscow Institute of Physics and Technology, Department of General and Applied Physics, Dolgoprudnyi, Russia a
We propose a new approach for decoding commands for EEG-based Brain–Computer Interface (BCI). Widely known BCI systems such as Berlin BCI and Graz BCI are based on the recognition of EEG patterns generated during imaginary motion or attention directed to different body parts. When the subject voluntarily performs the mental task, the corresponding EEG pattern can be recognized and decoded into the command for a technical device. Thus the most important part of such a BCI is the classifier of EEG patterns. The classifier of the Berlin BCI uses the method of Common Spatial Pattern (CSP), which appeared to be very efficient for separation of EEG patterns belonging to two classes corresponding to different mental states. The CSP algorithm finds directions in the N-dimensional space (where N is the number of EEG electrodes) that maximize the variance of projections onto these directions for one class and minimize the variance of projections for the other. The input for the Berlin BCI classifier is a vector of the largest and smallest variances along the found directions. This approach allows the diminishing of the intersection of the two classes. The input for the Graz BCI classifier is a vector of coefficients of autoregressive models of the third order created for electrical potentials recorded by all electrodes and between all pairs of electrodes. Thus the dimensionality of the input vector is 3N(N + 1) / 2. The recognition of the classes to which the input vectors belong is performed by a Bayesian classifier, assuming that for every class the input vectors are normally distributed. The first method demonstrates high efficiency while not requiring high computational resources, but it can be used only for discrimination of pairs of classes (i.e. the BCI produces only two commands). The second method allows the discrimination of many classes, but it requires much higher computational resources to achieve the same efficiency that could be crucial for real-time control.
181
Our approach combines advantages of both methods. It is based on Bayesian classification performed on multi-channel EEG recordings. Its efficiency was compared to the Berlin and Graz BCI in ten subjects in a simple computer game. The subjects imagined a motionless state or motion of one of the four limbs which gives five BCI commands. The suggested approach is shown to loose classifying accuracy only slightly when compared to traditional approaches, but allows the recognition of five mental states using reasonable computational resources.
doi:10.1016/j.ijpsycho.2008.05.478 SYMPOSIUM 27: Neurobiology of Motion and Emotion Imagery Symposium Chair: Anne Schienle (Austria) Neural correlations of imagery and observation of body movements: The influence of a first- and a third-person perspective R. Stark a, S. Pilgramm b, B. Lorey b, K. Zentgraf b, J. Munzert b, D. Vaitl a University of Giessen, Bender Institute of Neuroimaging, Giessen, Germany b University of Giessen, Department of Psychology and Sport Science, Giessen, Germany
a
Within the last twenty years it became obvious that imagery as well as observation of motor actions increases the neural activity in wide spread neural network including motor related brain areas as well as somatosensory areas although no overt motor action is indeed executed. While the phenomenon is indisputable there are open questions about factors which modulate the neural activity in the motor related areas during imagery and observation of motor actions. Following functional considerations we hypothesized that a first-person perspective results in greater neural activation than a third-person perspective. We conducted two fMRI studies: in a first study subjects were instructed to imagine simple hand movements either from a first- or third-person perspective. Further, the actual hand position was either compatible or incompatible to the imagined movement. In a second study complex whole body movements were used as stimuli. Dancing novices and experts observed ballroom dance videos. Hereby we could study the influence of expertise. Further, the videos show the dancing movements either from a first-person perspective (via helmet camera) or from a third-person perspective. Both studies replicate the previous findings that the imagination and the observation of motor movements yield neural activation in motor associated areas and somatosensory areas (e. g. supplementary motor area, secondary somatosensory area). More important it could be shown that according to our hypotheses a first-person perspective yielded more neural activity in the networks of interest than the third-person perspective. Especially the egocentric perspective displayed stronger activations in secondary somatosensory areas, in the inferior parietal cortex and in the insula.
doi:10.1016/j.ijpsycho.2008.05.479 Event-related EEG characteristics during motor imagery C. Neuper a, R. Scherer b, P. Grieshofer c, G. Pfurtscheller b University of Graz, Department of Psychology, Graz, Austria b Graz University of Technology, Institute for Knowledge Discovery, Graz, Austria c Rehabilitation Clinic Judendorf-Strassengel, Austria
a
Mental imagery of motor behaviour plays an important role in motor skill learning and in recovery of motor abilities after neurological damage. Brain imaging studies have revealed that motor imagery, defined as ‘the imagined rehearsal of a motor act without overt movements’, involves to a large extent the same cortical areas that are activated during actual motor preparation and execution. Accordingly, motor imagery produces changes in sensorimotor brain oscillations (i.e., event-related desynchronization (ERD) of mu and central beta rhythms) that occur naturally during movement. Since motor imagery shares physiological characteristics with movement execution, it represents an efficient mental strategy to operate a direct brain–computer interface (BCI). For this application, i.e., the control of an external device based on electrophysiological brain signals, it is essential
Spatial gradient of P300 in the brain–computer-interface paradigm I.A. Shevelev a, E.S. Mikhailova a, V.A. Chicherov a, V.A. Konishev b, D.V. Karlovskiy b a Institute of Higher Nervous Activity and Neurophysiology, Sensory Physiology Department, Moscow, Russia b Neurobotics Co, Zelenograd, Russia An analysis of the 2-D spatial gradient of P300 square that has an influence on the Brain–Computer-Interface (BCI) effectivity was performed in 16 healthy adult subjects. In the first experiment we studied recognition of different intended Cyrillic letters and words by P300 wave in the VEP. The set of optimal characteristics of visual stimulation by the letter matrix that provided maximal recognition level, as well as the best electrode position (Pz) were determined. It was found that the most effective criteria for ВСI decision were: P300 square, P300 amplitude and covariation coefficient. In the second experiment we measured the square of P300 wave for different location of the intended letters of the matrix. The horizontal and vertical spatial gradients of this square as a function of the angular distance between intended and the other letters and a tuning acuteness of visual attention to a significant letter were measured. We revealed the high acuteness of this tuning (width of the tuning curve at its half height was equal to 1.6grad.) independent of the letter position on the letter matrix. The horizontal and vertical gradients of P300 were found to be very similar, but in the half of cases the first one revealed some kind of the “lateral inhibition”: decrease of P300 square for the columns adjacent to the meaningful one. The tuning acuteness was found to be reliable and directly interrelated with P300 square. The data are evidenced high selectivity of local visual attention to letters and suggested some peculiarities of visual spatial attention to columns and rows of letters matrix.
doi:10.1016/j.ijpsycho.2008.05.477 Efficiency of three algorithms for classifying EEG patterns related to different mental tasks V.Y. Roschin a, M.S. Panko b, O.V. Ponfilenok b, A.V. Obukhov a Institute of Higher Nervous Activity and Neurophysiology RAS, Laboratory of Mathematical Neurobiology of Learning, Moscow, Russia b Moscow Institute of Physics and Technology, Department of General and Applied Physics, Dolgoprudnyi, Russia a
We propose a new approach for decoding commands for EEG-based Brain–Computer Interface (BCI). Widely known BCI systems such as Berlin BCI and Graz BCI are based on the recognition of EEG patterns generated during imaginary motion or attention directed to different body parts. When the subject voluntarily performs the mental task, the corresponding EEG pattern can be recognized and decoded into the command for a technical device. Thus the most important part of such a BCI is the classifier of EEG patterns. The classifier of the Berlin BCI uses the method of Common Spatial Pattern (CSP), which appeared to be very efficient for separation of EEG patterns belonging to two classes corresponding to different mental states. The CSP algorithm finds directions in the N-dimensional space (where N is the number of EEG electrodes) that maximize the variance of projections onto these directions for one class and minimize the variance of projections for the other. The input for the Berlin BCI classifier is a vector of the largest and smallest variances along the found directions. This approach allows the diminishing of the intersection of the two classes. The input for the Graz BCI classifier is a vector of coefficients of autoregressive models of the third order created for electrical potentials recorded by all electrodes and between all pairs of electrodes. Thus the dimensionality of the input vector is 3N(N + 1) / 2. The recognition of the classes to which the input vectors belong is performed by a Bayesian classifier, assuming that for every class the input vectors are normally distributed. The first method demonstrates high efficiency while not requiring high computational resources, but it can be used only for discrimination of pairs of classes (i.e. the BCI produces only two commands). The second method allows the discrimination of many classes, but it requires much higher computational resources to achieve the same efficiency that could be crucial for real-time control.
181
Our approach combines advantages of both methods. It is based on Bayesian classification performed on multi-channel EEG recordings. Its efficiency was compared to the Berlin and Graz BCI in ten subjects in a simple computer game. The subjects imagined a motionless state or motion of one of the four limbs which gives five BCI commands. The suggested approach is shown to loose classifying accuracy only slightly when compared to traditional approaches, but allows the recognition of five mental states using reasonable computational resources.
doi:10.1016/j.ijpsycho.2008.05.478 SYMPOSIUM 27: Neurobiology of Motion and Emotion Imagery Symposium Chair: Anne Schienle (Austria) Neural correlations of imagery and observation of body movements: The influence of a first- and a third-person perspective R. Stark a, S. Pilgramm b, B. Lorey b, K. Zentgraf b, J. Munzert b, D. Vaitl a University of Giessen, Bender Institute of Neuroimaging, Giessen, Germany b University of Giessen, Department of Psychology and Sport Science, Giessen, Germany
a
Within the last twenty years it became obvious that imagery as well as observation of motor actions increases the neural activity in wide spread neural network including motor related brain areas as well as somatosensory areas although no overt motor action is indeed executed. While the phenomenon is indisputable there are open questions about factors which modulate the neural activity in the motor related areas during imagery and observation of motor actions. Following functional considerations we hypothesized that a first-person perspective results in greater neural activation than a third-person perspective. We conducted two fMRI studies: in a first study subjects were instructed to imagine simple hand movements either from a first- or third-person perspective. Further, the actual hand position was either compatible or incompatible to the imagined movement. In a second study complex whole body movements were used as stimuli. Dancing novices and experts observed ballroom dance videos. Hereby we could study the influence of expertise. Further, the videos show the dancing movements either from a first-person perspective (via helmet camera) or from a third-person perspective. Both studies replicate the previous findings that the imagination and the observation of motor movements yield neural activation in motor associated areas and somatosensory areas (e. g. supplementary motor area, secondary somatosensory area). More important it could be shown that according to our hypotheses a first-person perspective yielded more neural activity in the networks of interest than the third-person perspective. Especially the egocentric perspective displayed stronger activations in secondary somatosensory areas, in the inferior parietal cortex and in the insula.
doi:10.1016/j.ijpsycho.2008.05.479 Event-related EEG characteristics during motor imagery C. Neuper a, R. Scherer b, P. Grieshofer c, G. Pfurtscheller b University of Graz, Department of Psychology, Graz, Austria b Graz University of Technology, Institute for Knowledge Discovery, Graz, Austria c Rehabilitation Clinic Judendorf-Strassengel, Austria
a
Mental imagery of motor behaviour plays an important role in motor skill learning and in recovery of motor abilities after neurological damage. Brain imaging studies have revealed that motor imagery, defined as ‘the imagined rehearsal of a motor act without overt movements’, involves to a large extent the same cortical areas that are activated during actual motor preparation and execution. Accordingly, motor imagery produces changes in sensorimotor brain oscillations (i.e., event-related desynchronization (ERD) of mu and central beta rhythms) that occur naturally during movement. Since motor imagery shares physiological characteristics with movement execution, it represents an efficient mental strategy to operate a direct brain–computer interface (BCI). For this application, i.e., the control of an external device based on electrophysiological brain signals, it is essential