Skip Navigation LinksHome > June 2010 - Volume 66 - Issue 6 > Finding Meaning in the Noise: Cortical Neuron Spike Predicti...
Neurosurgery:
doi: 10.1227/01.neu.0000375272.39513.69
Science Times

Finding Meaning in the Noise: Cortical Neuron Spike Prediction

ANDERSON, WILLIAM S.

Free Access

Because of the cellular complexity of the brain's cortex, underlying activity and intrinsic cellular wiring, the problem of understanding and predicting spiking patterns has seemed intractable. Truccolo et al (Truccolo W, Hochberg LR, Donoghue JP. Collective dynamics in human and monkey sensorimotor cortex: predicting single neuron spikes. Nature Neuroscience 2010;13(1):105–111), taking advantage of ongoing advances in microelectrode recording elements have now provided a general technique for demonstrating how single neuron spiking can be predicted by neighboring or distant cortical activity. This method relies on a point process model which can be fitted to a given neuron's spiking activity as a function of either the intrinsic cellular spiking history, or the spiking history of a collection of associated neurons. In a result that points out the rich dynamics of cortical interactions, a given neuron's spiking history was best predicted by the neuronal ensemble spiking history, and not by the prior spiking history of the neuron itself.

In this paper, Truccolo et al present a total of 12 data sets recorded from 2 human subjects and 4 monkeys in regions of the cortex controlling arm movements, including M1, parietal (Area 5d), and ventral premotor (PMv) cortex. In the humans, the recordings were made during motor control tasks in which the subjects used their own M1 spiking activity (recorded via the microelectrode array as a neuroprosthesis) to move a cursor on a computer screen. Recordings were made with a 10 * 10 array of silicon microelectrodes mounted on a 4.2 mm * 4.2 mm platform. Protrusion depths of the individual electrodes ranged from 1–1. mm. The system provides access to single unit and few-unit extracellular recordings from each individual electrode.

To reduce the calculational complexity of the problem of estimating the conditional probability for a single neuron's activity as influenced by the surrounding cellular ensemble, the authors approximate the conditional probability as the probed neuron's instantaneous spike rate multiplied by the time discretization step. This instantaneous spike rate was then modeled as the exponential of a sum of terms consisting of the individual spiking rates from the surrounding neuronal ensemble, weighted by a set of temporal filters. These temporal filters represent the fitting functions for the model, and can be separated into terms dependent on the intrinsic spiking history of the neuron being studied, and on terms dependent on the surrounding or separated influencing neuronal ensemble. A log-likelihood estimation procedure is then performed to derive the temporal filter function parameters as a fit to the time sequence data.

Using this method, the authors demonstrated excellent receiver operator curves for predicting single neuron spiking as determined by the full history model, or a model incorporating only the surrounding ensemble of cells (Figure). Similar results occurred in both human and monkey motor cortex. A predictive power estimate derived from the area under the curve of the ROC also demonstrated substantial results. Some normalized predictive powers had magnitudes greater than 1 implying a redundancy in the predictive powers of the intrinsic and aggregate cellular models. Perhaps most interesting in this paper was the result that the predictive power of inter-areal neuronal ensembles (as opposed to the neuronal ensemble at the site of the probed neuron) was also very high, highlighting the power of this technique to help unravel functional brain connections over wide areas.

FIGURE. A and B, rec...
FIGURE. A and B, rec...
Image Tools

In summary, Truccolo et al have provided a general method for demonstrating the predictive power of neuronal ensemble spiking histories on individual cortical neuronal activity. The method may prove to be too general for implementation in fast data acquisition systems, and it is not clear how the algorithm would perform given truly random or low information content input. However, it has direct relevance in the design of prosthetic systems that rely on cortical input, and could prove to be a useful tool in teasing apart the dynamic interactions of cortical regions.

WILLIAM S. ANDERSON

Copyright © by the Congress of Neurological Surgeons

Login

Article Tools

Images

Share