Spiros H. Courellis

Restoring lost cognitive function

A prosthetic device that functions in a biomimetic manner to replace information transmission between cortical brain regions is considered. In such a prosthesis, damaged CNS neurons is replaced with a biomimetic system comprised of silicon neurons. The replacement silicon neurons would have functional properties specific to those of the damaged neurons and would both receive as inputs and send as outputs electrical activity to regions of the brain with which the damaged region previously communicated. Thus, the class of prosthesis proposed is one that would replace the computational function of the damaged brain and restore the transmission of that computational result to other regions of the nervous system.

Nonlinear Modeling of Causal Interrelationships in Neuronal Ensembles

The increasing availability of multiunit recordings gives new urgency to the need for effective analysis of ldquomultidimensionalrdquo time-series data that are derived from the recorded activity of neuronal ensembles in the form of multiple sequences of action potentials-treated mathematically as point-processes and computationally as spike-trains. Whether in conditions of spontaneous activity or under conditions of external stimulation, the objective is the identification and quantification of possible causal links among the neurons generating the observed binary signals. A multiple-input/multiple-output (MIMO) modeling methodology is presented that can be used to quantify the neuronal dynamics of causal interrelationships in neuronal ensembles using spike-train data recorded from individual neurons. These causal interrelationships are modeled as transformations of spike-trains recorded from a set of neurons designated as the ldquoinputsrdquo into spike-trains recorded from another set of neurons designated as the ldquooutputsrdquo. The MIMO model is composed of a set of multiinput/single-output (MISO) modules, one for each output. Each module is the cascade of a MISO Volterra model and a threshold operator generating the output spikes. The Laguerre expansion approach is used to estimate the Volterra kernels of each MISO module from the respective input-output data using the least-squares method. The predictive performance of the model is evaluated with the use of the receiver operating characteristic (ROC) curve, from which the optimum threshold is also selected. The Mann-Whitney statistic is used to select the significant inputs for each output by examining the statistical significance of improvements in the predictive accuracy of the model when the respective inputs is included. Illustrative examples are presented for a simulated system and for an actual application using multiunit data recordings from the hippocampus of a behaving rat.

A biosensor for detecting changes in cognitive processing based on nonlinear systems analysis.

A new type of biosensor, based on hippocampal slices cultured on multielectrode arrays, and using nonlinear systems analysis for the detection and classification of agents interfering with cognitive function is described. A new method for calculating first and second order kernel was applied for impulse input-spike output datasets and results are presented to show the reliability of the estimations of this parameter. We further decomposed second order kernels as a sum of nine exponentially decaying Laguerre base functions. The data indicate that the method also reliably estimates these nine parameters. Thus, the state of the system can now be described with a set of ten parameters (first order kernel plus nine coefficients of Laguerre base functions) that can be used for detection and classification purposes.

An efficient method for studying short-term plasticity with random impulse train stimuli

In this article, we introduce an efficient method that models quantitatively nonlinear dynamics associated with short-term plasticity (STP) in biological neural systems. It is based on the Voterra-Wiener modeling approach adapted for special stimulus/response datasets. The stimuli are random impulse trains (RITs) of fixed amplitude and Poisson distributed, variable interimpulse intervals. The class of stimuli, we use can be viewed as a hybrid between the paired impulse approach (variable interimpulse interval between two input impulses) and the fixed frequency approach (impulses repeated at fixed intervals, varying in frequency from one stimulus dataset to the next). The responses are sequences of population spike amplitudes of variable size and are assumed to be contemporaneous with the corresponding impulses in the RITs they are evoked by. The nonlinear dynamics of the mechanisms underlying STP are captured by kernels used to create compact STP models with predictive capabilities. Compared to similar methods in the literature, the method presented in this article provides a comprehensive model of STP with considerable improvement in prediction accuracy and requires shorter experimental data collection time.

Detection and classification of neurotoxins using a novel short-term plasticity quantification method

A tissue-based biosensor is described for screening chemical compounds that rapidly affect the nervous system. The proposed sensor is an extension of a previous work on cultured hippocampal slices [Biosens. Bioelectron. 16 (2001) 491]. The detection of the chemical compounds is based on a novel quantification method of short-term plasticity (STP) of the CA1 system in acute hippocampal slices, using random electrical impulse sequences as inputs and population spike (PS) amplitudes as outputs. STP is quantified by the first and the second order kernels using a variant of the Volterra modeling approach. This approach is more specific and time-efficient than the conventional paired pulse and fixed frequency train methods [J. Neurosci. Methods 2 (2002) 111]. Describing the functional state of the biosensor, the kernels changed accordingly as chemical compounds were added. The second order kernel was decomposed into nine Laguerre functions. The corresponding Laguerre coefficients along with the first order kernel were used as features for classification purposes. The biosensor was tested using picrotoxin (100 microM), trimethylopropane phosphate (10 microM), tetraethylammonium (4 mM), valproate (5 mM), carbachol (5 mM), DAP5 (25 microM), CNQX (3 microM), and DNQX (0.15, 1.5, 3, 5 and 10 microM). Each chemical compound gave a different feature profile corresponding to its pharmacological class. The first order kernel and the Laguerre coefficients formed the input to an artificial neural network (ANN) comprised of a single layer of perceptrons. The ANN was able to classify each tested compound into its respective class.

An algorithm for real-time extraction of population EPSP and population spike amplitudes from hippocampal field potential recordings

A new method is presented for extracting the amplitude of excitatory post synaptic potentials (EPSPs) and spikes in real time. It includes a low pass filter (LPF), a differentiator, a threshold function, and an intelligent integrator. It was applied to EPSP and population spike data recorded in the Dentate Gyrus and the CA1 hippocampus in vitro. The accuracy of the extraction algorithm was evaluated via the extraction normalized mean square error (eNMSE) and was found to be very high (eNMSE < 5%). The preservation of neuronal information was confirmed using the Volterra-Poisson modeling approach. Volterra-Poisson kernels were computed using amplitudes extracted with both proposed and traditional methods. The accuracy of the computed kernels and the resulting model was evaluated via the prediction normalized mean square error (pNMSE) and was found to be very high (pNMSE < 5%). The similarity between the kernels computed when the proposed method was used to extract the field potential amplitude and their counterparts when the traditional method was used to extract the field potential amplitude confirms the preservation of the neuronal dynamics. The proposed method represents a new class of real time field potential amplitude extraction algorithms with complexity that can be included in hardware implementations.

An artificial neural network for motion detection and speed estimation

An artificial neural network with nonlinear spatiotemporal features performing motion detection (i.e. speed estimation and direction selection) is introduced. The input to the network is a moving visual pattern, and the output is the direction of motion (directional selectivity) and an estimate of the velocity magnitude (speed estimation). To carry out the process, the network utilizes three layers. The first layer performs spatiotemporal processing using a difference-of-Gaussians spatial filter and a bandpass differentiating temporal filter. In the second layer, nonlinear spatiotemporal operations extract features pertinent to the estimation of speed-in the form of spikelike trains-and directional selectivity. The third layer provides connections and nodes where directional selectivity is implemented, and an estimate of the speed is formed by temporally processing the spikelike trains provided by the second layer. Computer simulations of a one-dimensional application of the network are presented, illustrating the responses to moving spots and edges with a number of velocity profiles. The fundamental limitations on the proposed network are explored as well

Nonlinear Dynamic Model of CA1 Short-Term Plasticity using Random Impulse Train Stimulation

A comprehensive, quantitative description of the nonlinear dynamic characteristics of the short-term plasticity (STP) in the CA1 hippocampal region is presented. It is based on the Volterra–Poisson modeling approach using random impulse train (RIT) stimuli. In vitro hippocampal slice preparations were used from adult rats. RIT stimuli were applied at the Schaffer collaterals and population spike responses were recorded at the CA1 cell body layer. The computed STP descriptors that capture the nonlinear dynamics of the underlying STP mechanisms were the Volterra–Poisson kernels. The kernels quantified the presence of facilitatory and inhibitory STP behavior in magnitude and duration. A third order Volterra–Poisson STP model was introduced that accurately predicted in-sample and out-of-sample system responses. The proposed model could also accurately predict impulse pair and short impulse train system responses.

Principal dynamic mode analysis of action potential firing in a spider mechanoreceptor

The encoding of mechanical stimuli into action potentials in two types of spider mechanoreceptor neurons is modeled by use of the principal dynamic modes (PDM) methodology. The PDM model is equivalent to the general Wiener–Bose model and consists of a minimum set of linear dynamic filters (PDMs), followed by a multivariate static nonlinearity and a threshold function. The PDMs are obtained by performing eigen-decomposition of a matrix constructed using the first-order and second-order Volterra kernels of the system, which are estimated by means of the Laguerre expansion technique, utilizing measurements of pseudorandom mechanical stimulation (input signal) and the resulting action potentials (output signal). The static nonlinearity, which can be viewed as a measure of the probability of action potential firing as a function of the PDM output values, is computed as the locus of points of the latter that correspond to output action potentials. The performance of the model is assessed by computing receiver operating characteristic (ROC) curves, akin to the ones used in decision theory and quantified by computing the area under the ROC curve. Three PDMs are revealed by the analysis. The first PDM exhibits a high-pass characteristic, illustrating the importance of the velocity of slit displacement in the generation of action potentials at the mechanoreceptor output, while the second and third PDMs exhibit band-pass and low-pass characteristics, respectively. The corresponding three-input nonlinearity exhibits asymmetric behavior with respect to its arguments, suggesting directional dependence of the mechanoreceptor response on the mechanical stimulation and the PDM outputs, in agreement to our findings from a previous study (Ann Biomed Eng 27:391–402, 1999). Differences between the Type A and B neurons are observed in the zeroth-order Volterra kernels (related to the average firing), as well as in the magnitudes of the second and third PDMs that perform band-pass and low-pass processing of the input signal, respectively.

A multi-input modeling approach to quantify hippocampal nonlinear dynamic transformations

A multi-input modeling approach is introduced to quantify hippocampal neural dynamics. It is based on the Volterra modeling approach extended to multiple inputs. The computed Volterra kernels allow quantitative description of hippocampal transformations and define a predictive model that can produce responses to arbitrary input patterns. Electrophysiological data from several CA3 and CA1 cells in behaving rats were recorded simultaneously using an array of penetrating electrodes. This activity was used to compute kernels up to third order for single and multiple input cases. Representative sets of kernels illustrate the variability of the dynamics of the CA3-CA1 transformations. Our models predictive accuracy was evaluated using ROC curves