|
|
|
|
Neuroinformatics: ZonMw Building, Den Haag, Friday, November 26, 2004 |
| Abstracts - Lectures |
|
Three case studies for NEURON
The simulation environment NEURON (developed by Michael Hines and available in the public domain) is widely used for computational neuroscience problems at the membrane, cell and local network level. There is a large database that contains most of the membrane mechanisms that determine neuronal excitability (ion channels, pumps, receptors) and the Internet contains databases that can provide real cell morphologies of neurons of many different brain regions. In this talk three case studies will be shown in order to provide an idea of the type of problems that are well handled by NEURON. First we will show how "peculiar" experimental voltage-clamp data of calcium currents can be explained if we assume that during development low voltage activated calcium channels move from the soma into the dendritic tree. In the second example the interaction between neurons, glia and the interstitial space is modeled in order to analyse the generation of epileptic seizures and spreading depression. In the last example we will generate a hypothesis for homeostatic control of neuronal excitability using the Ih potassium current.
Neuroinformatics: bridging the gap between neuron and neuro-imaging
One of the main challenges for the future is to bridge the gap between the single and multiple neuron activity and the neuronal interactions at the one hand, and neuroimaging signals at the other hand. In particular, it is crucially important to interpret EEG/MEG data (for example the initiation and disappearance of beta or gamma-rhythms) in relation to specific cognitive activities in terms of interactions between neurons. In order to bridge this gap, at least two parallel approaches should be taken. The first (experimental) approach would the simultaneous recording of single-unit activity, multi-unit activity, local field potentials and EEG/MEG data. The second (theoretical) approach is to develop good models that capture the characteristic properties of realistic neurons and their interactions. Such models are crucial to predict the evolution of the behavior of a network of neurons, and how their contribution to EEG/MEG signals will change under various conditions. As such they will provide a tool to test hypotheses about information processing at the level of the neuron in human subjects by means of EEG/MEG. In this presentation we will provide an overview of the state of the art regarding the experimental and theoretical approach, as well as an overview of future steps to be taken.
Non-classical receptive field inhibition and contour detection
Various visual effects show that the perception of an edge or line can be influenced by other such stimuli in the surroundings. Such effects can be related to non-classical receptive field (non-CRF) inhibition that is found in 80% of the orientation selective neurons in the primary visual cortex. A mathematical model of non-CRF inhibition is presented in which the response of an orientation selective cell is suppressed by the responses of other such cells beyond its classical (excitatory) receptive field. Non-CRF inhibition acts as a feature contrast computation for oriented stimuli: the response to an optimal stimulus over the receptive field is suppressed by similar stimuli in the surround. Consequently, it strongly reduces the responses to texture edges while scarcely affecting the responses to isolated contours. The biological utility of this neural mechanism might thus be that of contour (vs. texture) detection. The results of computer simulations based on the proposed model explain perceptual effects, such as orientation contrast pop-out, 'social conformity' of lines embedded in gratings, reduced saliency of contours surrounded by textures and decreased visibility of letters embedded in band-limited noise [Petkov and Westenberg, 2003 Biological Cybernetics 88: 236-246]. The insights into the biological role of non-CRF inhibition can be utilized in machine vision. The proposed model is employed in a contour detection algorithm that substantially outperforms previously known such algorithms in computer vision [Grigorescu et al, 2003 IEEE Transactions on Image Processing 12 729-739] [Grigorescu et al, 2004 Image and Vision Computing 22 609-622]. A demonstration will be given of a Gabor filter augmented with surround inhibition of texture edges. (The operator is available on-line at www.cs.rug.nl/~petkov).
Literature (downloadable from http://www.cs.rug.nl/~petkov/publications/journals.html):
Presynaptic mechanisms of synaptic plasticity
Synapses are not static but their performance is modified adaptively in response to activity. This might have functional consequences for information processing at the network level. We study presynaptic mechanisms that determine this plasticity with the ultimate goal to construct a realistic model for neurotransmitter release that can be implemented in neural network models. Model parameters are obtained from electrophysiological measurements from cultured neurons that make synaptic contacts on themselves (autapses) allowing to measure synaptic plasticity in a controlled manner. In these cells presynaptic proteins are manipulated by means of viral overexpression or the use of transgenic animals in order to investigate their role in neurotransmitter release and synaptic plasticity.
Computational Analysis of Spatiotemporal Patterns of Activity in Neuronal Networks
Information processing in the brain is based on spatiotemporal patterns of electrical activity in neuronal networks. New experimental techniques allow the monitoring of these patterns in great detail by simultaneously recording neuronal activity from a large number of network locations. To be able to analyze the flood of data these techniques produce, we are developing (1) statistical methods for analyzing spatiotemporal patterns of neuronal activity and (2) computational models of neuronal networks, both macroscopic networks and neuronal microcircuits, to simulate these patterns and understand them in relation to structural and functional connectivity. The models are validated with the extensive data we have on spatiotemporal patterns in cortical brain slices and cultured neuronal networks, and will be used in our search for key genetic regulators of network activity and animal behavior.
Color constancy and coding strategies of horizontal cells in the retina
The purpose of this project is to clarify the role of the horizontal cell feedback circuit in the coding of information in the vertebrate retina, in particular the retina of zebrafish. It is part of our ongoing research program which strives to: 1) understand neural processing of natural stimuli and 2) develop a quantitative understanding of the first processing steps in the visual system. In the project, the statistics of visual stimuli as normally encountered by the retina will be investigated, and used to investigate the coding of information at the photoreceptor-horizontal cell level. Naturalistic stimuli will be presented to the eye for developing good models both of the function and the physiology of the circuit. The final goal is to obtain a quantitative understanding of the signal coding schemes in the outer retina.
Modeling, analysis and visualization of brain connectivity
In het onderzoek op het gebied van functionele neuroimaging (fMRI, PET, EEG) heeft lange tijd een sterk accent gelegen op het vinden van actieve gebieden ("hot spots") in het brein als functie van sensorische stimuli of cognitieve taken. Recentelijk is er meer aandacht gekomen voor bestudering de relaties tussen de verschillende hersengebieden en de dynamica van hersenprocessen. Een belangrijk concept is dat van de "brain connectivity", die een brug kan slaan tussen de opvattingen over functionele segregatie vs. integratie. De modellering, analyse en visualisatie van deze netwerken vormt een uitdagend probleem. Recente ontwikkelingen en onderzoeksplannen op dit gebied zullen kort worden besproken.
Role of attention in reinforcement learning
Here we propose a new role for feedback connections in learning. Our aim is to understand the neuronal plasticity that underlies animal learning in classification tasks. Our new model, called AGREL (attention-gated reinforcement learning), is designed in accordance with neurobiological principles, as are previous reinforcement learning schemes. The new scheme is a major advancement over previous models, however, since we show that feedback connections can guide the formation of sensory representations that are useful for the task at hand.
Simultaneous EEG and fMRI for the localisation of spontaneous alpha-rhythm
Recently, the possibility has been described in the literature to register EEG during simultaneous fMRI scanning. In principle, this new technique poses the opporunity to localize the underlying generators of classical EEG phenomena, such as the alpha-rhythm.
In theory, the principle is quite simple. By using the waxing and waning of the alpha rhythm (or the presence/absence of another EEG phenomenon) as a reference function, one can detect the fMRI pixels with a high correlation to the EEG derived reference.
In practice, however, there are technical problems to be solved in the field of time series analysis: the artefact removal in the EEG, the demodulation of the EEG, the extraction of an EEG reference from the multi-channels, the determination of a sensible correlation co-efficient and its statistical significance.
In my presentation I would like to address these technical issues and show our preliminary results.
Neuroinformatics of neuronal network dynamics
Brain function is based on the electrical activity in neuronal networks, and their understanding is one of the major challenges in neuroscience research. Ongoing developments in both computational and experimental techniques contribute strongly to the progress in this field. Neural simulators become more powerful and increase the scale and detail of investigation. Multielectrode techniques make it possible to record action potential firing activity from increasing numbers of nerve cells.
These developments now result in a great need for (i) analytical tools to quantify and analyse the large amount of information in the experimental data, (ii) database tools to make experimental data available to experimental and computational investigators, and (iii) computational models to help understanding the system from which the data has been recorded. Without doubt one can state that neuronal network research will profit strongly from the Neuroinformatics objectives with respect to data basing and data sharing, analytical tool development and computational modeling.
A neuroinformatics system integrating experimental data, analytical tools and computational models on neuronal network dynamics is for that reason greatly needed.
Computer vision algorithms for segmentation and recognition in computer-aided
diagnosis
Abstract 1
Abstract 2
The neuroinformatics PhD program in Edinburgh
|
|
Abstracts - Posters |
|
Gabor filtering augmented with surround inhibition for improved contour detection by texture suppression
Gabor functions are adequate models of simple cell receptive fields [Daugman, 1985 J. Opt. Soc. Am. A 2 1160-1169] [Jones and Palmer, 1987 J. Neurophysiology 58 1233-1258]. Gabor-energy filters have been used to model complex cells [Morrone and Burr, 1988 Proc. Royal Society London B 235 221-245]. Such filters augmented with non-classical receptive field inhibition [Nothdurft et al, 1999 Visual Neuroscience 16 15-34] were demonstrated to mimic various perceptual effects [Petkov and Westenberg, 2003 Biological Cybernetics 88 236-246]. Furthermore, they substantially outperform previously known contour detection algorithms [Grigorescu et al, 2003 IEEE Transactions on Image Processing 12 729-739]. Nowadays Gabor filters are widely used in various scientific fields: visual perception, computational neuroscience, image processing and computer vision. We implemented a Gabor filter augmented with surround inhibition and made it available on internet [http://www.cs.rug.nl/~imaging, http://www.cs.rug.nl/~petkov] for use by other researchers. Users can choose from the image material available on the site or upload their own images. The following parameters can be selected: preferred wavelenght; preferred orientation; phase offset; aspect ratio; bandwidth; number of orientations; half-wave rectification threshold; superposition of results for different phase offsets; type, strength and area of surround inhibition; post-processing. The site includes a number of visual perception examples.
Non-classical receptive field inhibition and its relation to orientation-contrast pop-out and line and contour saliency - a computational approach
Non-classical receptive field (non-CRF) inhibition has been suggested as the possible origin of various perceptual effects, such as overestimation of an acute angle between two lines [Blakemore et al., 1970 Nature 228 37-39] and orientation-contrast pop-out [Knierim and van Essen, 1992 J. Neurophysiology 67 961-980] [Nothdurft et al, 1999 Visual Neuroscience 16 15-34]. Computational models of two types of cell that incorporate non-CRF inhibition, which are based on Gabor energy filters extended by surround suppression of two kinds, isotropic and anisotropic, were introduced in [Petkov and Westenberg, 2003 Biological Cybernetics 88 236-246]. We apply these computational models to the images used to demonstrate the referred perceptual effects. The results of these computer simulations confirm the rightness of the hypothesis for a possible functional role of non-CRF inhibition in the referred and further effects, such as reduced saliency of lines and contours embedded in gratings [Galli and Zama, 1931 Zeitschrift für Psychologie 31 308-348] [Kanizsa, 1979 Organization in Vision, Essays on Gestalt Perception (New York: Praeger)] and reduced saliency of contours surrounded by textures. We made the algorithms and images available on internet [http://www.cs.rug.nl/~imaging] for use by other researchers. A demonstration will be given on site.
Effects of periodic single-site voltage stimulations on spontaneous network
bursting
Neural networks cultured on multielectrode arrays are characterised by synchronous firing periods (bursts) among neurons (for review see Corner et al. 2002). We have recently shown that in dissociated rat neocortical networks, profiles of spontaneously occurring network bursts change with the development of the culture (Van Pelt et al., 2004 a, b). Among others, the width of averaged bursts showed significant changes during development and went through a broadening and shortening phase in about 4 weeks. To investigate the influence of electrical stimulation on the profile of spontaneous network bursts, we measured spontaneous activity before and after stimulation and analysed the network bursts as previously reported for experiments without stimulation (Van Pelt et al., 2004 a, b). Neurons from E18 Wistar rats were dissociated and plated on multielectrode arrays (60 electrodes with diameters of either 10, 20 or 30 µm). Culture density was approximately 100 000 cells / plate. Measurements of extracellular spikes were performed on cultures of 2-4 weeks in vitro. The following stimulation protocol was applied sequentially at 4-6 sites: a train of 20-40 biphasic voltage pulses of either ± 2V or ± 0.3V was delivered. Stimulation frequency varied and was between 0.1 Hz and 0. 5Hz. We found, that electrical stimulation not only changes the temporal profile of the averaged spontaneous bursts, but can also silence or recruit sites which contribute to these network bursts. We conclude that stimulation can change the routing of spontaneous activity through the network.
References:
Recent approaches to the study of ongoing neuronal oscillations
Neuronal networks generate complex patterns of activity even in the absence of sensory input [1]. This 'ongoing activity', often in the form of oscillations, plays a critical role in development, reflects functional and structural organization of the brain, and can bias conscious perception of sensory stimuli [2].
The unprecedented computational power and rapid advances in digital signal processing have rejuvenated the study of ongoing oscillations, allowing us to extract new information out of what was previously considered noise. I will present recently developed frameworks and data-analysis approaches to understanding the character and function of ongoing neuronal oscillations.
1. Linkenkaer-Hansen et al., J Neurosci 21:1370-1377, 2001.
2. Linkenkaer-Hansen et al., J Neurosci (in press), 2004.
High speed two-photon imaging of calcium rise time kinetics and calcium handling in dendritic spines
Rise times of calcium transients in dendritic spines may be crucial for synaptic plasticity. With the use of ultra-fast two-photon point imaging we were able to measure both rise- and decay kinetics of fast calcium transients in spines and small dendrites of pyramidal cells. We observed that both rise and decay kinetics are at least twice as fast in spines compared to the parent dendrite. To investigate whether differences in calcium kinetics between spines and dendrites were due to morphological differences in surface to volume ratio (SVR) we constructed a one-compartment dynamical model to simulate fast calcium signals. Numerical simulations showed that morphological parameters alone are not sufficient to explain differences in calcium handling of dendrites and spines. The model predicts that spines and small dendrites differ in endogenous buffer capacity. In addition, it predicts differential dynamics of free calcium and calmodulin-CAMKII activation in these compartments.
Computational Analysis of Spatiotemporal Patterns of Activity in Neuronal Networks
Abstract - see lecture
Transmission of population codes through layered networks
We discuss feed-forward architectures that compute with population codes. Using radial basis functions we implement a layered network of noisy integrate-and-fire neurons which computes the sum of two population coded quantities. The network performs the computation robustly, accurately and quickly.
|
|
|
|
|