Predicting locomotor activity via neural ensemble recordings in the primary motor cortex in rats.
The possibility to use information from cortical neurons to drive neuroprosthetic devices is an area that has recently garnered a lot of attention within neuroscience. Most of this research has been directed towards restoring upper limb motility. In this study, we have started to assess the possibility of using cortical neurons to drive a lower limb neuroprosthetic device. Rats were chronically implanted with multielectrode arrays (consisting of between 32-96 electrodes wires) in the primary motor cortex. Neuronal activity (both spikes and local field potentials) were recorded while the animals walked on a treadmill. Joint and limb positions were captured using UV marks placed on the animal and captured in 3D at 100Hz using four high-speed cameras and motion capture software. An artificial model of the rats movement was then reconstructed to use as a comparison for the neuronal model. Spikes were sorted offline and different units were shown to be significantly responsive to locomotor activity. These spikes were binned and used as regressors in a linear filter, with joint positions modeled as a weighted linear combination of neuronal activity using multidimensional linear regression. Preliminary performances of the linear models suggest that the cortical activity can predict the actual motor activity well (up to R2 = 0.64). This indicates that there is some correlation between higher cortical activity in MI and locomotor activity. It also indicates that cortical activity may be useful for controlling a neuroprosthetic or robotic device, at even a low-level of control.