Spiking neural networks Ghosh-Dastidar, Samanwoy; Adeli, Hojjat
International journal of neural systems,
08/2009, Letnik:
19, Številka:
4
Journal Article
Recenzirano
Most current Artificial Neural Network (ANN) models are based on highly simplified brain dynamics. They have been used as powerful computational tools to solve complex pattern recognition, function ...estimation, and classification problems. ANNs have been evolving towards more powerful and more biologically realistic models. In the past decade, Spiking Neural Networks (SNNs) have been developed which comprise of spiking neurons. Information transfer in these neurons mimics the information transfer in biological neurons, i.e., via the precise timing of spikes or a sequence of spikes. To facilitate learning in such networks, new learning algorithms based on varying degrees of biological plausibility have also been developed recently. Addition of the temporal dimension for information encoding in SNNs yields new insight into the dynamics of the human brain and could result in compact representations of large neural networks. As such, SNNs have great potential for solving complicated time-dependent pattern recognition problems because of their inherent dynamic representation. This article presents a state-of-the-art review of the development of spiking neurons and SNNs, and provides insight into their evolution as the third generation neural networks.
A novel wavelet-chaos-neural network methodology is presented for classification of electroencephalograms (EEGs) into healthy, ictal, and interictal EEGs. Wavelet analysis is used to decompose the ...EEG into delta, theta, alpha, beta, and gamma sub-bands. Three parameters are employed for EEG representation: standard deviation (quantifying the signal variance), correlation dimension, and largest Lyapunov exponent (quantifying the non-linear chaotic dynamics of the signal). The classification accuracies of the following techniques are compared: 1) unsupervised-means clustering; 2) linear and quadratic discriminant analysis; 3) radial basis function neural network; 4) Levenberg-Marquardt backpropagation neural network (LMBPNN). To reduce the computing time and output analysis, the research was performed in two phases: band-specific analysis and mixed-band analysis. In phase two, over 500 different combinations of mixed-band feature spaces consisting of promising parameters from phase one of the research were investigated. It is concluded that all three key components of the wavelet-chaos-neural network methodology are important for improving the EEG classification accuracy. Judicious combinations of parameters and classifiers are needed to accurately discriminate between the three types of EEGs. It was discovered that a particular mixed-band feature space consisting of nine parameters and LMBPNN result in the highest classification accuracy, a high value of 96.7%.
A spatio-temporal wavelet-chaos methodology is presented for analysis of EEGs and their
delta,
theta,
alpha, and
beta sub-bands for discovering potential markers of abnormality in Alzheimer's disease ...(AD). The non-linear dynamics of the EEG and EEG sub-bands are quantified in the form of the correlation dimension (CD), representing system complexity, and the largest Lyapunov exponent (LLE), representing system chaoticity. The methodology is applied to two groups of EEGs: healthy subjects and AD patients. The eyes open and eyes closed conditions are investigated to evaluate the effect of visual input and attention. EEGs from different loci in the brain are investigated to discover areas of the brain responsible for or affected by changes in CD and LLE. It is found that the wavelet-chaos methodology and the sub-band analysis developed in this research accurately characterizes the non-linear dynamics of non-stationary EEG-like signals with respect to the EEG complexity and chaoticity. It is concluded that changes in the brain dynamics are not spread out equally across the spectrum of the EEG and over the entire brain, but are localized to certain frequency bands and electrode loci. New potential markers of abnormality were discovered in this research for both eyes open and closed conditions.
Abstract Background Extremely high accuracy for predicting CT+ traumatic brain injury (TBI) using a quantitative EEG (QEEG) based multivariate classification algorithm was demonstrated in an ...independent validation trial, in Emergency Department (ED) patients, using an easy to use handheld device. This study compares the predictive power using that algorithm (which includes LOC and amnesia), to the predictive power of LOC alone or LOC plus traumatic amnesia. Participants ED patients 18–85 years presenting within 72 h of closed head injury, with GSC 12–15, were study candidates. 680 patients with known absence or presence of LOC were enrolled (145 CT + and 535 CT − patients). Methods 5–10 min of eyes closed EEG was acquired using the Ahead 300 handheld device, from frontal and frontotemporal regions. The same classification algorithm methodology was used for both the EEG based and the LOC based algorithms. Predictive power was evaluated using area under the ROC curve (AUC) and odds ratios. Results The QEEG based classification algorithm demonstrated significant improvement in predictive power compared with LOC alone, both in improved AUC (83% improvement) and odds ratio (increase from 4.65 to 16.22). Adding RGA and/or PTA to LOC was not improved over LOC alone. Conclusions Rapid triage of TBI relies on strong initial predictors. Addition of an electrophysiological based marker was shown to outperform report of LOC alone or LOC plus amnesia, in determining risk of an intracranial bleed. In addition, ease of use at point-of-care, non-invasive, and rapid result using such technology suggests significant value added to standard clinical prediction.
A new
Multi-Spiking Neural Network (
MuSpiNN) model is presented in which information from one neuron is transmitted to the next in the form of multiple spikes via multiple synapses. A new supervised ...learning algorithm, dubbed
Multi-SpikeProp, is developed for training MuSpiNN. The model and learning algorithm employ the heuristic rules and optimum parameter values presented by the authors in a recent paper that improved the efficiency of the original single-spiking Spiking Neural Network (SNN) model by two orders of magnitude. The classification accuracies of MuSpiNN and Multi-SpikeProp are evaluated using three increasingly more complicated problems: the XOR problem, the Fisher iris classification problem, and the epilepsy and seizure detection (EEG classification) problem. It is observed that MuSpiNN learns the XOR problem in twice the number of epochs compared with the single-spiking SNN model but requires only one-fourth the number of synapses. For the iris and EEG classification problems, a modular architecture is employed to reduce each 3-class classification problem to three 2-class classification problems and improve the classification accuracy. For the complicated EEG classification problem a classification accuracy in the range of 90.7%–94.8% was achieved, which is significantly higher than the 82% classification accuracy obtained using the single-spiking SNN with SpikeProp.
Abstract Background There is an urgent need for objective criteria adjunctive to standard clinical assessment of acute Traumatic Brain Injury (TBI). Details of the development of a quantitative index ...to identify structural brain injury based on brain electrical activity will be described. Methods Acute closed head injured and normal patients ( n =1470) were recruited from 16 US Emergency Departments and evaluated using brain electrical activity (EEG) recorded from forehead electrodes. Patients had high GCS (median=15), and most presented with low suspicion of brain injury. Patients were divided into a CT positive (CT+) group and a group with CT negative findings or where CT scans were not ordered according to standard assessment (CT−/CT_NR). Three different classifier methodologies, Ensemble Harmony, Least Absolute Shrinkage and Selection Operator (LASSO), and Genetic Algorithm (GA), were utilized. Results Similar performance accuracy was obtained for all three methodologies with an average sensitivity/specificity of 97.5%/59.5%, area under the curves (AUC) of 0.90 and average Negative Predictive Validity (NPV)>99%. Sensitivity was highest for CT+ cases with potentially life threatening hematomas, where two of three classifiers were 100%. Conclusion Similar performance of these classifiers suggests that the optimal separation of the populations was obtained given the overlap of the underlying distributions of features of brain activity. High sensitivity to CT+ injuries (highest in hematomas) and specificity significantly higher than that obtained using ED guidelines for imaging, supports the enhanced clinical utility of this technology and suggests the potential role in the objective, rapid and more optimal triage of TBI patients.
: An improved freeway incident‐detection model is presented based on speed, volume, and occupancy data from a single detector station using a combination of wavelet‐based signal processing, ...statistical cluster analysis, and neural network pattern recognition. A comparative study of different wavelets (Haar, second‐order Daubechies, and second‐ and fourth‐order Coifman wavelets) and filtering schemes is conducted in terms of efficacy and accuracy of smoothing. It is concluded that the fourth‐order Coifman wavelet is more effective than other types of wavelets for the traffic incident detection problem. A statistical multivariate analysis based on the Mahalanobis distance is employed to perform data clustering and parameter reduction to reduce the size of the input space for the subsequent step of classification by the Levenberg–Marquardt backpropagation (BP) neural network. For a straight two‐lane freeway using real data, the model yields an incident detection rate of 100%, false alarm rate of 0.3%, and detection time of 35.6 seconds.
The goal of this research is to develop an efficient SNN model for epilepsy and epileptic seizure detection using electroencephalograms (EEGs), a complicated pattern recognition problem. Three ...training algorithms are investigated: SpikeProp (using both incremental and batch processing), QuickProp, and RProp. Since the epilepsy and epileptic seizure detection problem requires a large training dataset the efficacy of these algorithms is investigated by first applying them to the XOR and Fisher iris benchmark problems. Three measures of performance are investigated: number of convergence epochs, computational efficiency, and classification accuracy. Extensive parametric analysis is performed to identify heuristic rules and optimum parameter values that increase the computational efficiency and classification accuracy. The result is a remarkable increase in computational efficiency. For the XOR problem, the computational efficiency of SpikeProp, QuickProp, and RProp is increased by a factor of 588, 82, and 75, respectively, compared with the results reported in the literature. EEGs from three different subject groups are analyzed: (a) healthy subjects, (b) epileptic subjects during a seizure-free interval, and (c) epileptic subjects during a seizure. It is concluded that RProp is the best training algorithm because it has the highest classification accuracy among all training algorithms specially for large size training datasets with about the same computational efficiency provided by SpikeProp. The SNN model for EEG classification and epilepsy and seizure detection uses RProp as training algorithm. This model yields a high classification accuracy of 92.5%.
Prediction or early-stage diagnosis of Alzheimer's disease (AD) requires a comprehensive understanding of the underlying mechanisms of the disease and its progression. Researchers in this area have ...approached the problem from multiple directions by attempting to develop (a) neurological (neurobiological and neurochemical) models, (b) analytical models for anatomical and functional brain images, (c) analytical feature extraction models for electroencephalograms (EEGs), (d) classification models for positive identification of AD, and (e) neural models of memory and memory impairment in AD. This article presents a state-of-the-art review of research performed on computational modeling of AD and its markers. The review covers the following approaches: computer imaging, classification models, connectionist neural models, and biophysical neural models. It is concluded that a mixture of markers and a combination of novel computational techniques such as neural computing, chaos theory, and wavelets can increase the accuracy of algorithms for automated detection and diagnosis of AD.