| Brain machine interfaces (BMIs) aim to provide new rehabilitation options and channels of interaction to patients who have lost their ability to move their limbs due to disease or injury to the central or peripheral nervous system. Brain activity produces a variety of electrical signals that can be measured through diverse recording technologies and are potential candidates for BMI inputs. Electrocorticogram recordings (ECoG) provide an intermediate level of abstraction between invasive microarray recordings that penetrate the tissue and noninvasive scalp recordings (EEG). In order to map ECoG activity to motor behavior, extraction of the appropriate spatiotemporal and spectral control features is critical. Though they provide higher amplitude, less noisy and broader band signals compared to EEG recordings, extraction of signatures of motor events in spontaneous ECoG activity still entails challenges and remains unexplored. Herein, clinical behavioral paradigms are developed to record and analyze very broadband ECoG from two epileptic patients participating in reaching, pointing and cursor tracking tasks. Although historically frequencies above the high gamma band have been discarded as background activity, we study all frequencies upto the Nyquist frequency (of 6.1kHz) by dividing the broadband into logarithmically equal bands, yielding passbands of equal center frequency to bandwidth ratio. The role of the spectral resolution, in which the broadband is partitioned, in the reconstruction of the patients' hand trajectories is studied through crosscorrelations, event related synchronizations, directional tuning, and source separation methodologies. Mapping of neural modulations to goal-oriented motor behavior is achieved via the traditionally used linear adaptive filters and as a novelty through nonlinear echo state networks with upto 85% correlation. Regularization methodologies for online feature selection and subspace projection through semiblind source separation algorithms are implemented for further reduction of the vast feature space. Removal of interictal spiking activity present across the sensorimotor cortex of the patients, which supresses the motor control features, is studied through source separation. The reconstructed hand trajectories are analyzed through spatial and spectral sensitivity. Presence of motor features up to 6kHz is shown as a novel contribution in the field of ECoG BMIs. |
