Simon Benjaminsson

A Novel Model-Free Data Analysis Technique Based on Clustering in a Mutual Information Space : Application to Resting-State fMRI

Non-parametric data-driven analysis techniques can be used to study datasets with few assumptions about the data and underlying experiment. Variations of independent component analysis (ICA) have been the methods mostly used on fMRI data, e.g., in finding resting-state networks thought to reflect the connectivity of the brain. Here we present a novel data analysis technique and demonstrate it on resting-state fMRI data. It is a generic method with few underlying assumptions about the data. The results are built from the statistical relations between all input voxels, resulting in a whole-brain analysis on a voxel level. It has good scalability properties and the parallel implementation is capable of handling large datasets and databases. From the mutual information between the activities of the voxels over time, a distance matrix is created for all voxels in the input space. Multidimensional scaling is used to put the voxels in a lower-dimensional space reflecting the dependency relations based on the distance matrix. By performing clustering in this space we can find the strong statistical regularities in the data, which for the resting-state data turns out to be the resting-state networks. The decomposition is performed in the last step of the algorithm and is computationally simple. This opens up for rapid analysis and visualization of the data on different spatial levels, as well as automatically finding a suitable number of decomposition components.

Arterial input function derived from pairwise correlations between PET-image voxels

A metabolite corrected arterial input function is a prerequisite for quantification of positron emission tomography (PET) data by compartmental analysis. This quantitative approach is also necessary for radioligands without suitable reference regions in brain. The measurement is laborious and requires cannulation of a peripheral artery, a procedure that can be associated with patient discomfort and potential adverse events. A non invasive procedure for obtaining the arterial input function is thus preferable. In this study, we present a novel method to obtain image-derived input functions (IDIFs). The method is based on calculation of the Pearson correlation coefficient between the time-activity curves of voxel pairs in the PET image to localize voxels displaying blood-like behavior. The method was evaluated using data obtained in human studies with the radioligands [ 11 C]flumazenil and [ 11 C]AZ10419369, and its performance was compared with three previously published methods. The distribution volumes (VT) obtained using IDIFs were compared with those obtained using traditional arterial measurements. Overall, the agreement in VT was good (~3% difference) for input functions obtained using the pairwise correlation approach. This approach performed similarly or even better than the other methods, and could be considered in applied clinical studies. Applications to other radioligands are needed for further verification.

From ANN to Biomimetic Information Processing

Artificial neural networks (ANN) are useful components in today’s data analysis toolbox. They were initially inspired by the brain but are today accepted to be quite different from it. ANN typically lack scalability and mostly rely on supervised learning, both of which are biologically implausible features. Here we describe and evaluate a novel cortex-inspired hybrid algorithm. It is found to perform on par with a Support Vector Machine (SVM) in classification of activation patterns from the rat olfactory bulb. On-line unsupervised learning is shown to provide significant tolerance to sensor drift, an important property of algorithms used to analyze chemo-sensor data. Scalability of the approach is illustrated on the MNIST dataset of handwritten digits.

A large-scale model of the three first stages of the mammalian olfactory system implemented with spiking neurons

In this work we present a large-scale three stage model of the early mammalian olfactory system, including the olfactory epithelium (OE), the olfactory bulb (OB) and the olfactory (piriform) cortex (OC). All neurons in the network are modeled with a single or few compartments using the Hodgkin-Huxley formalism and are implemented in the NEURON simulator for parallel execution [1,2]. We investigate the dynamics of the network response to odorants and its performance in odor classification experiments. The OE model comprises families of olfactory receptor neurons (ORNs) with different sensitivities, each family expressing one type of olfactory receptor (OR) with a vector of affinity values for each ligand [3]. These different ORN families connect to distinct glomeruli and mitral cells (MT) according to a hypothesized wiring scheme to form a fuzzy interval code for odorant concentration in the OB, i.e. each MT cell responds within a certain range of odorant concentration and these ranges overlap for different MT cells within one glomerulus. Mitral cells in different glomeruli respond independently to an odourant forming a sparse and distributed code in the OB, i.e. only a fraction of MT cells in different glomeruli is active when an odorant is present. The OC is modeled by a modular attractor network of pyramidal cells and inhibitory interneurons [4]. The OB response patterns are used to self-organize a projection to the OC based on learning algorithms employing ideas from machine learning (mutual information between the MT responses, multi-dimensional scaling, vector quantization and Hebbian learning) [5,6]. As a result, odourant responses are represented by a sparse distributed code in the OC. Results from runs with network sizes comprising thousands of model neurons show that this biophysically plausible network model generates response patterns of cells reminding of their real counterparts (see Figure ​Figure1),1), produces attractor dynamics in the olfactory cortex, and is able to discriminate between the different trained odors. We investigate effects of the model size, backprojections from OC to OB, study the performance of the model in discriminating mixtures of odorants and compare Calcium concentration in the olfactory cortex with experimental measurements [7]. Figure 1 Schematic of the model (A) and sample membrane potentials of a pyramidal cell (B), a mitral cell (C) and an olfactory receptor neuron (D)

Nexa: A scalable neural simulator with integrated analysis

Large-scale neural simulations encompass challenges in simulator design, data handling and understanding of simulation output. As the computational power of supercomputers and the size of network models increase, these challenges become even more pronounced. Here we introduce the experimental scalable neural simulator Nexa, for parallel simulation of large-scale neural network models at a high level of biological abstraction and for exploration of the simulation methods involved. It includes firing-rate models and capabilities to build networks using machine learning inspired methods for e.g. self-organization of network architecture and for structural plasticity. We show scalability up to the size of the largest machines currently available for a number of model scenarios. We further demonstrate simulator integration with online analysis and real-time visualization as scalable solutions for the data handling challenges.

PT2 Reverse Engineering of Nature in the Field of Chemical Sensors

A large scale chemical sensor array consisting of 16384 conducting polymer elements was developed emulating characteristics of the biological olfactory receptor system. A biologically realistic computational model of the olfactory cortex was developed and the data from the large array was used to test the performance of the system. Classification of odorants and segmentation of mixtures of were investigated and the results were compared to that from support vector machine algorithms.

Odor recognition in an attractor network model of the mammalian olfactory cortex

Odor recognition constitutes a key functional aspect of olfaction and in real-world scenarios it requires that odorants occurring in complex chemical mixtures are identified irrespective of their concentrations. We investigate this challenging pattern recognition problem in the framework of a three-stage computational model of the mammalian olfactory system. To this end, we first synthesize odor stimuli with the primary representations in the olfactory receptor neuron (ORN) layer and the secondary representations in the output of the olfactory bulb (OB) in the model. Next, sparse olfactory codes are extracted and fed into the recurrent network model, where as a result of Hebbian-like associative learning an attractor memory storage is produced. We demonstrate the capability of the resultant olfactory cortex (OC) model to perform robust odor recognition tasks and offer computational insights into the underlying network mechanisms.

Odor segmentation and identification in an abstract large-scale model of the mammalian olfactory system.

We integrate an olfactory epithelium and bulb (OB) model with an olfactory (piriform) cortex (OC) model, all comprising abstract graded units representing local populations of neurons, to demonstrate the overall systems applicability to odor recognition tasks. Large-scale simulations on a cluster computer allow us to evaluate an instance of the model with the size matching that of the olfactory system of a small mammal. The input data is synthesized with the intention to resemble the distribution of various features of early olfactory response patterns to naturalistic odor stimuli, e.g. [1,2]. Self-organization of the feedforward connectivity from the OB to the piriform cortex based on statistical properties of synthetic olfactory stimuli with the support of synaptic plasticity provide the capacity for generating sparse and distributed cortical representations [3]. At the cortical level, associative functions resulting from recurrent connectivity with Hebbian plasticity implement attractor memory storage and pattern processing, e.g. completion and rivalry [4]. Adaptation in cortical units, mimicking neuronal spike frequency adaptation, serves as an underlying mechanism for odor segmentation. We also investigate how backprojecting connections from the OC to the OB facilitate context dependent stimulus identification. The model is evaluated in real-world scenarios involving odor classification and segmentation in noisy environments. Firstly, the network is trained to perform concentration invariant odor recognition. Secondly, the model identifies single-ligand components of odor mixtures it is subjected to in a testing phase. Finally, the effect of context dependence is evaluated by examining the systems capability to recognize specific low intensity stimuli when it is forced into an arousal state. Preliminary results show the dependence of performance on the network size, the number of odors and the complexity of their early olfactory activation patterns. In a broader context, the proposed large-scale abstract models of holistic olfactory processing implements a multiscale approach to modeling olfaction. After the network has been trained, its connection matrices are utilized in simulating a more biophysically realistic model of the mammalian olfactory system (for details see poster by Kaplan et al.).