Uncovering multiscale neural network dynamics underlying reach-to-grasp movements
Abstract
We will develop a novel multiscale network dynamical modeling framework for spike-field activity and use it to investigate neural dynamics underlying motor behavior at multiple spatial and temporal scales. Neural representations span various scales from single neurons to large neuronal networks within and across brain regions. Thus, to fully understand the neural basis of behavior, neural dynamics need to be studied simultaneously at these scales. Current technology enables measuring the brain at multiple scales by concurrently recording spikes from individual neurons and field activities~such as the electrocorticogram (ECoG) and local field potential (LFP)~, which capture aggregate network processes across thousands of neurons. How neuralnetwork dynamics at different spatial and temporal scales give rise to motor behavior, the relationship between these scales, and their causal interactions are poorly understood. Thus, critical to our understanding of motor cortex is investigating neural dynamics at multiple spatial and temporal scales simultaneously by modeling high-dimensional spike-field network dynamics,which has not been possible to date. This investigation has been impeded by two major challenges. First, multiscale dynamical modeling is computationally challenging due to the high-dimension and the fundamentally different time-scales and statistical profiles of spikes and fields. Spikes are binary-valued and have a fast millisecond time-scale while fields are continuous-valued and haveslower time-scales. Second, the dimension and nature of neural dynamics likely depend on task complexity, thus necessitating the recording of spike-field network activity during naturalistic 3- dimensional (3D) motor tasks that go beyond simpler 2D tasks used in the vast majority of prior studies. In this program, we will address these two challenges: (1) We will develop multiscalemachine learning algorithms to identify a low-dimensional hidden dynamical state for high dimensional hybrid spike-field network activity. This dynamical state will summarize key multiscale activity patterns that generate the movement. We will also develop multiscale causality measures, which are currently lacking, to examine the flow of information in spike-field networks and reconstruct their functional topology. (2) We will use these computational tools to characterizemultiscale spike-field dynamics and uncover causal multiscale interactions across premotor and primary motor cortical areas in a complex naturalistic 3D reach-to-grasp task in which all 27 degree-of-freedom of the upper-extremity are monitored in non-human primates. This program will advance our understanding of neural control of movement and provide new computationaltools to investigate multiscale neural dynamics in general. The set of tools and the new motor control insights can inform the design of future robotic systems, human-computer interface technologies, and neural-inspired information processing systems of interest to DoD.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 25, 2019
- Source ID
- N000141912128
Entities
People
- Maryam Shanechi
Organizations
- Office of Naval Research
- United States Navy
- University of Southern California