Charles S. DaSalla

2009 Special Issue: Single-trial classification of vowel speech imagery using common spatial patterns

With the goal of providing a speech prosthesis for individuals with severe communication impairments, we propose a control scheme for brain-computer interfaces using vowel speech imagery. Electroencephalography was recorded in three healthy subjects for three tasks, imaginary speech of the English vowels /a/ and /u/, and a no action state as control. Trial averages revealed readiness potentials at 200 ms after stimulus and speech related potentials peaking after 350 ms. Spatial filters optimized for task discrimination were designed using the common spatial patterns method, and the resultant feature vectors were classified using a nonlinear support vector machine. Overall classification accuracies ranged from 68% to 78%. Results indicate significant potential for the use of vowel speech imagery as a speech prosthesis controller.

Reconstruction of flexor and extensor muscle activities from electroencephalography cortical currents.

The ability to reconstruct muscle activity time series from electroencephalography (EEG) may lead to drastic improvements in brain-machine interfaces (BMIs) by providing a means for realistic continuous reproduction of dexterous movements in human beings. However, it is considered difficult to isolate signals related to individual muscle activities from EEG because EEG sensors record a mixture of signals originating from many cortical regions. Here, we challenge this assumption by reconstructing agonist and antagonist muscle activities (i.e. filtered electromyography (EMG) signals) from EEG cortical currents estimated using a hierarchical Bayesian EEG inverse method. Results of 5 volunteer subjects performing isometric right wrist flexion and extension tasks showed that individual muscle activity time series, as well as muscle activities at different force levels, were well reconstructed using EEG cortical currents and with significantly higher accuracy than when directly reconstructing from EEG sensor signals. Moreover, spatial distribution of weight values for reconstruction models revealed that highly contributing cortical sources to flexion and extension tasks were mutually exclusive, even though they were mapped onto the same cortical region. These results suggest that EEG sensor signals were reasonably isolated into cortical currents using the applied method and provide the first evidence that agonist and antagonist muscle activity time series can be reconstructed using EEG cortical currents.

Neuroanatomical correlates of brain–computer interface performance

Brain-computer interfaces (BCIs) offer a potential means to replace or restore lost motor function. However, BCI performance varies considerably between users, the reasons for which are poorly understood. Here we investigated the relationship between sensorimotor rhythm (SMR)-based BCI performance and brain structure. Participants were instructed to control a computer cursor using right- and left-hand motor imagery, which primarily modulated their left- and right-hemispheric SMR powers, respectively. Although most participants were able to control the BCI with success rates significantly above chance level even at the first encounter, they also showed substantial inter-individual variability in BCI success rate. Participants also underwent T1-weighted three-dimensional structural magnetic resonance imaging (MRI). The MRI data were subjected to voxel-based morphometry using BCI success rate as an independent variable. We found that BCI performance correlated with gray matter volume of the supplementary motor area, supplementary somatosensory area, and dorsal premotor cortex. We suggest that SMR-based BCI performance is associated with development of non-primary somatosensory and motor areas. Advancing our understanding of BCI performance in relation to its neuroanatomical correlates may lead to better customization of BCIs based on individual brain structure.

Spatial filtering and single-trial classification of EEG during vowel speech imagery

With the purpose of providing assistive technology for the communication impaired, we propose a control algorithm for speech prostheses using vowel speech imagery. Electroen-cephalograms were recorded in three healthy subjects during the performance of three tasks, imaginary speech of the English vowels /a/ and /u/, and a no action state as control. Speech related potentials were visualized by grand averaging in the time domain. Feature data was obtained by filtering the time series data using optimal spatial filters designed through the common spatial patterns method. Resultant feature vectors were classified using a nonlinear support vector machine. Overall classification accuracies ranged from 68 to 78%. Results indicate significant potential for the use of vowel speech imagery as a speech prosthesis controller.

Usability of EEG cortical currents in classification of vowel speech imagery

With the purpose of providing assistive technology for the communication impaired, we propose a new approach for speech prostheses using vowel speech imagery. Using a hierarchical Bayesian method, electroencephalography (EEG) cortical currents were estimated using EEG signals recorded from three healthy subjects during the performance of three tasks, imaginary speech of vowels /a/ and /u/, and a no imagery state as control. The 3-task classification using a sparse logistic regression method with variational approximation (SLR-VAR) revealed that mean classification accuracy of cortical currents was almost two times greater than chance level and significantly higher than that using EEG signals. The results suggest the possibility of using EEG cortical currents to discriminate multiple syllables by improving the spatial discrimination of EEG.

Dissociable neural representations of wrist motor coordinate frames in human motor cortices

There is a growing interest in how the brain transforms body part positioning in the extrinsic environment into an intrinsic coordinate frame during motor control. To explore the human brain areas representing intrinsic and extrinsic coordinate frames, this fMRI study examined neural representation of motor cortices while human participants performed isometric wrist flexions and extensions in different forearm postures, thereby applying the same wrist actions (representing the intrinsic coordinate frame) to different movement directions (representing the extrinsic coordinate frame). Using sparse logistic regression, critical voxels involving pattern information that specifically discriminates wrist action (flexion vs. extension) and movement direction (upward vs. downward) were identified within the primary motor and premotor cortices. Analyses of classifier weights further identified contributions of the primary motor cortex to the intrinsic coordinate frame and the ventral and dorsal premotor cortex and supplementary motor area proper to the extrinsic coordinate frame. These results are consistent with existing findings using non-human primates and demonstrate the distributed representations of independent coordinate frames in the human brain.

Mastery Biases Agent-Representation in Visual Perception of Handwritings

This study investigated whether normal adults were able to discriminate agency from the perceived traces or trajectories of past actions such as handwritings. Subjects wrote two types of component parts of Chinese characters, either mastered and unmastered, and were later shown various handwritten strokes and judged whether each of them had been written by themselves or by someone else. We found that people tended to answer that the handwritings had been written by others when they saw unmastered types of strokes, while they tended to answer that the handwritings had been written by themselves when they saw mastered types of strokes. This finding suggests a tight interplay among perception, self-consciousness, and memorized action in the motor system and adds to our knowledge about a higher order representation level in the agency recognition. Possible cognitive neuroscientific implications and engineering applications of the finding are also discussed.

Classifying vowel speech imagery using EEG cortical currents

regression, called smooth sparse regression, which has a spatio-temporal “smoothing” prior that encourages weights that are close in time and space to have similar values. This causes the classifier to select spatio-temporally continuous groups of features, whereas standard sparse regression classifiers often select a scattered collection of independent features. We applied the proposed method to both simulation data and real MEG data from two separate experiments. We found that the method consistently increases classification accuracy, and produces weight vectors that are more meaningful from a neuroscientific perspective. Research fund: JST PRESTO.