Determining Patterns in Neural Activity for Reaching Movements Using Nonnegative Matrix Factorization
- PDF / 1,235,392 Bytes
- 9 Pages / 600 x 792 pts Page_size
- 96 Downloads / 220 Views
Determining Patterns in Neural Activity for Reaching Movements Using Nonnegative Matrix Factorization Sung-Phil Kim Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32611, USA Email: [email protected]
Yadunandana N. Rao Motorola Inc., FL, USA Email: [email protected]
Deniz Erdogmus Department of Computer Science and Biomedical Engineering, Oregon Health & Science University, Beaverton, OR 97006, USA Email: [email protected]
Justin C. Sanchez Department of Pediatrics, Division of Neurology, University of Florida, Gainesville, FL 32611, USA Email: [email protected]
Miguel A. L. Nicolelis Department of Neurobiology, Center for Neuroengineering, Duke University, Durham, NC 27710, USA Emails: [email protected]; [email protected]
Jose C. Principe Department of Electrical and Computer Engineering, University of Florida, Gainesville, FL 32611, USA Email: [email protected] Received 31 January 2004; Revised 23 March 2005 We propose the use of nonnegative matrix factorization (NMF) as a model-independent methodology to analyze neural activity. We demonstrate that, using this technique, it is possible to identify local spatiotemporal patterns of neural activity in the form of sparse basis vectors. In addition, the sparseness of these bases can help infer correlations between cortical firing patterns and behavior. We demonstrate the utility of this approach using neural recordings collected in a brain-machine interface (BMI) setting. The results indicate that, using the NMF analysis, it is possible to improve the performance of BMI models through appropriate pruning of inputs. Keywords and phrases: brain-machine interfaces, nonnegative matrix factorization, spatiotemporal patterns, neural firing activity.
1.
INTRODUCTION
Brain-machine interfaces (BMIs) are an emerging field that aims at directly transferring the subject’s intent of movement to an external machine. Our goal is to engineer devices that are able to interpret neural activity originating in the motor cortex and generate accurate predictions of hand position. In the BMI experimental paradigm, hundreds of microelectrodes are implanted in the premotor, motor, and
posterior parietal areas and the corresponding neural activity is recorded synchronously with behavior (hand reaching and grasping movements). Spike detection and sorting algorithms are used to determine the firing times of single neurons. Typically, the spike-time information is summarized into bin counts using short windows (100 milliseconds in this paper). A number of laboratories including our own have demonstrated that linear and nonlinear adaptive system identification approaches using the bin count input
3114
EURASIP Journal on Applied Signal Processing
can lead to BMIs that effectively predict the hand position and grasping force of primates for different movement tasks [1, 2, 3, 4, 5, 6, 7, 8]. The adaptive methods studied thus far include moving average models, time-delay neural networks (TDNNs), Kalman filter and extensions, recursive m
Data Loading...
