Neural Prosthetics: Continuous Cortical Control of a Robotic Arm for Self Feeding (Dec 2005 to Feb 2008)
During the first part of my Phd, I worked on the development of the next generation of brain computer interfaces (BCI).
In this project, we have demonstrated the feasibility of using neural data from the motor-cortex to control a physical device with impressive precision.
We have developed a state-of-the-art BCI system, which performed extraction and decoding of neural signals enabling real-time control of an external 4 degrees-of-freedom robotic arm with 2 fingers.
An example of a monkey using the BCI for self feeding:
Dynamic Functional Connectivity Between Cortex and Muscles (Feb 2008 to April 2012)
During the second part of my Phd, I investigated how the motor cortex controls reach-to-grasp movements, with a focus on wrist and fingers movement.
Better understanding of how the brain controls natural movement would enable us to build better neural prosthetics controlling a large number of degrees-of-freedom.
A basic estimate of static functional-connectivity can be obtained using correlation. The interactive visualization below shows the correlation between 92 motor cortex neurons and 16 muscles.
Small nodes represent neurons, large nodes represent muscles. Muscle colors vary from black (proximal arm muscles) to dark gray (wrist muscles) and light gray (distal finger muscles).
Each link color represents the mean correlation value between the neuron and muscle across a variety of behaviors. Only statistically significant correlation values outside of the range [-0.2,0.2] were used in this animation to make visualization easier (thus some neuron nodes seem to be floating). Hover over a node to see its label. Drag nodes to view the network structure from different views.
Try to figure out which muscle group tends to show higher correlation values and what muscles across groups seem to be similarly correlation to the same neurons.
Encoding of Movement Parameters By Different Neural Modalities (May 2012 to Present)
Different signal modalities recorded from the brain contain varying amounts of movemnet related information. We record single-unit, multi-unit and local-field potential activity simultaneously with hand kinematics, and model differences in movement representation across modalities.
Example of movement related LFP activity from a single channel in the frequency domain is shown in the figure below.
Movment related activity can be clearly seen around 230ms, the beginning of movement: lower-frequency bands show a decrease in activity (blue values), whether higher-frequency bands show increased activity (yellow-red colors).
Click the image to enlarge.
Multi-Modality BCI Decoding (Nov 2012 to Present)
Intracortical local-field potentials (LFP) are a potential information source for brain-compute-interfaces, which could replace or supplement Multi-Unit Activity (MUA). We study LFP tuning during motor-tasks to inform us how LFP might be used for decoding movement. A variety of decoding algorithms with different mixtures of LFP and MUA are used to determine the best usage for LFP in BCI.