Detection of Epileptiform Activity in Human EEG Signals Using Bayesian Neural Networks

Size: px
Start display at page:

Download "Detection of Epileptiform Activity in Human EEG Signals Using Bayesian Neural Networks"

Transcription

1 Neural Information Processing - Letters and Reviews Vol.0, No., January 006 LETTER Detection of Epileptiform Activity in Human EEG Signals Using Bayesian Neural Networks School of Electrical and Information Engineering University of the Witwatersrand, Johannesburg Private Bag 3, Wits 050, South Africa nadm@absmail.co.za, d.rubin@ee.wit.ac.za, tmarwala@yahoo.com (Submitted on September 5, 005; Accepted on January 4, 006) Abstract We investigate the application of neural networks to the problem of detecting interictal epileptiform activity in the electroencephalogram (EEG). The proposed detector consists of a segmentation, feature extraction and classification stage. For the feature extraction stage, coefficients of the Discrete Wavelet Transform (DWT), real and imaginary parts of the Fast Fourier Transform and raw EEG data were all found to be wellsuited to EEG classification. Principal Component Analysis was used to reduce the dimensionality of the features. For the classification stage, Multi-Layer Perceptron neural networks were implemented according to Maximum Likelihood and Bayesian Learning formulations. The latter was found to make better use of training data and consequently produced better trained neural networks. Rejection thresholds of 0.9 were applied to the network output as a doubt level in order to ensure that only reliable classification decisions are made. A maximum classifier accuracy of 95.% was achieved with 5.0% of patterns not being classified. Bayesian moderated outputs could not improve on these classification predictions significantly enough to warrant their added computational overhead. Keywords Epilepsy, Artificial neural network, Multi-layer Perceptron, Maximum Likelihood. Introduction Epilepsy is a common neurological disorder that is primarily diagnosed and monitored using the electroencephalogram (EEG). The EEG is a temporal recording of the variations in the weak electric potentials that develop across the scalp. Epileptiform activity is manifested in the EEG as transient waveforms known as spikes and sharp waves (SSWs). Figure shows examples of the some of these waveforms. Despite advances in technology, analysis of EEG records is still usually performed manually by highly trained professionals who visually screen the EEG for the presence of spike waveforms. This is a laborious task as the EEG with its multiple recording channels can produce large volumes of data. Partially or fully automating this task with a computer-based pattern recognition system will result in a significant time saving. Research into techniques for automatic classification of the EEG began as early as the 960s and has produced a variety of different approaches to solving this problem. Earlier systems employed mimetic, rulesbased [, ] or arithmetic [3] detectors. These have the advantage of being relatively simple to implement, however, suffer from high false detection rates. This is mostly caused by the inter-subject variability between the EEGs of different patients due to factors such as age, level of consciousness and inability to control all experimental conditions. Adaptive techniques which are more tolerant of variations in input signals are expected to be better suited to EEG classification than classical methods because of the wide inter-individual variation. This expectation has led to numerous applications of adaptive techniques, in particular the Artificial Neural Network (ANN), to EEG classification [4, 5, 6]. These studies generally report better classification results than studies that use alternative methods.

2 Detection of Epileptiform Activity in Human EEG Signals Figure. Examples of (A) epileptic spikes, (B) sharp waves and (C) spike and wave complex in the EEG. The ANN is a powerful data processing system consisting of a large number highly interconnected, smaller processing units called neurons. Their structure is loosely based on the structure of biological neurons in the human brain and gives them the ability to learn a classification function from example training data. The learning attribute allows for the implementation of a detector based on example data instead of specifying a complete set of detection rules or logic. This is an attractive property when classifying complex signals such as EEG signals whose variability between patients and even within the same patient does not make it feasible to specify a single, comprehensive set of rules for classification. In this study, the application of ANNs to the problem of epileptic SSW detection is further studied. In particular, we apply state-of-the-art, principled techniques for neural network training in order to optimise results. The problem of neural network training is posed in a Maximum Likelihood and Bayesian Learning formulation. Specifically, the error function and output activation function of the ANN are chosen so that the output of ANN approximates the posterior probability of class membership as derived by Bishop [7]. Rejection thresholds or doubt levels can then be applied to this output in order for classification decisions to only be made when there is a high degree of certainty in the class membership of an input pattern. It has been argued [8] that this will lead to classifiers that are perceived as being more reliable as there will be a lower number of false detections and less need for manual correction. Applying Bayesian inference techniques to neural network training and prediction offers principled methods for determining optimal weight decay coefficients and model selection while making efficient use of training data. This is useful for problems where only a limited set of data is available for training and testing of a neural network. The suitability both of Bayesian and Maximum Likelihood ANNs for EEG SSW detection is assessed in this study... Epileptic and Non-Epileptic Waveforms. EEG Waveforms Abnormal EEG activity associated with epilepsy may be ictal (during seizures) or inter-ictal (between seizures) []. In this study, only inter-ictal activity is considered as this is more commonly used for clinical diagnosis of epilepsy. The following types of waveforms fall under this category: ) Spikes and Sharp Waves (SSWs): These are paroxysmal, transient waveforms. They have high amplitudes relative to background activity and pointed peaks. The spikes have duration of 0 70 ms and sharp waves have duration of ms. Polyspikes is the term given to multiple repetitive spikes that are repeated at a frequency of approximately 0Hz.

3 Neural Information Processing - Letters and Reviews Vol.0, No., January 006 ) Spike and Wave Complex: SSWs are sometimes followed by a slow wave of frequency less than 4 Hz. Numerous artifacts are also often present in the EEG that is not generated by cerebral activity. These include eye-blink artifacts, eye movement artifacts, and Electromyographic (EMG) muscular artifacts as well as electrical interference. Some of these signals such as eye-blink artifacts bear a close resemblance to SSWs. The design of a detection system should facilitate a distinction between SSWs and artifacts. 3. Neural Networks for Classification For the purposes of this study, ANNs are viewed as a general framework for representing non-linear mappings between multi-dimensional spaces, where the form of mapping is governed by a set of adjustable parameters [7]. With supervised learning, the training data consists of both the input to the ANN and an associated target output. In this study, the input is a set of features extracted from the EEG and the target output is or 0, where represents the epileptiform activity pattern class and 0 represents the non-epileptiform activity pattern class. 3.. Architecture The Multi-Layer Perceptron (MLP) ANN architecture was selected for this study. This architecture has been applied successfully to the epileptic SSW detection problem in the past [4, 5, 6], which suggests that it is suitable for classifying the EEG. Every connection between inputs and neurons is weighted by an adjustable weight parameter. Theoretically, this type of two-layer MLP model is able to approximate any continuous function with arbitrary accuracy, provided the number of hidden neurons is sufficiently large [7]. If x is the input to the MLP and y is the output of the MLP, a relation mapping between the input to the output may be written as M N y = f output w j fhidden wij xi + b0 + w0 () j= i= where N is the number of inputs units, M is the number of hidden neurons, x i is the i-th input unit, w ij is the weight parameter between input i and hidden neuron j, and w j is the weight parameter between hidden neuron j and the output neuron. The activation function f ( ) is sigmoid as output f + e The activation function f ( ) is hyperbolic tangent as hidden 3.. Maximum Likelihood Formulation output ( a) = () a f ( a) = tan h( a) (3) hidden For a two-class classification problem, Bishop [7] showed that the error between the network output in Eq.() and the target output t for all training patterns P is the cross-entropy error function given by E P [ t p y p ) + ( t p )ln( y p ] CEE = ) p= ln( (4) During training, the ANN weight parameters w s are iteratively adjusted in order to minimise the error function in Eq.(4). Initially the weights are set to random values. An optimisation method known as the scaled conjugate gradient (SCG) method [9] was used to determine the magnitude of the parameter updates. The SCG method was selected instead of other optimization methods because of its fast convergence properties. Bishop [7] proved that if the cross-entropy error function is minimized during ANN training and the activation function of an ANN is given by Eq.(), the output of an ANN approximates the posterior probability of membership to a pattern class given the vector of inputs x. In the case of EEG SSW detection, the output will approximate the posterior probability of the input EEG pattern containing epileptic activity. If this class is represented by C and the pattern class for not containing epileptic activity is represented by C, the relations for the posterior probability of class membership can be written as P ( C x) = y (5) 3

4 Detection of Epileptiform Activity in Human EEG Signals P( C x) = y (6) The above relations give a probabilistic interpretation to the ANN output. Based on these relations, it is clear that an input vector has a high probability of belonging to class C when y is close to, and C when y is close to 0. If y is close to 0.5, there is uncertainty in the class membership of the input vector. A simple method to ensure that the classifier makes classification decisions only where there is a high degree of certainty is to apply an upper and lower rejection threshold to the ANN output as proposed by Bishop [7]. This classification decision rule is defined as Decide C if y > θ; decide C if y < (-θ); (7) Otherwise do not classify x The parameter θ sets the level of the rejection threshold. 3.3 Bayesian Formulation The maximum likelihood approach to neural network training assumes that weight parameters have fixed unknown values and estimates these values by minimizing a suitable error function. However, as the vector of weight parameters w are estimated from a finite set of data, there is always some degree of uncertainty associated with the value of this weight vector. For example, if a single training pattern of data is removed from the training set, the weight vector that is learnt during training will differ from the weight vector that is learnt when the same training pattern is present in the training set. Bayesian learning takes into account the uncertainty in the weight vector by assigning a probability distribution to the weight vector that represents the relative degrees of belief in the values of the weight vector. An MLP model structure reflecting the number of hidden layer neurons is first defined as H, and a prior distribution p(w H) is assigned to the weight vector to reflect any initial beliefs regarding the distribution of the weight parameters. Once a set of training data, D, is observed, the prior distribution can be converted to a posterior distribution of network weights using the Bayes theorem as p( D w, H ) p( w H ) p( w D, H ) = (8) p( D H ) p(d w, H) is the likelihood function and p(d H) is the evidence of model H. MacKay [0] uses a Gaussian prior as α = p(w H ) exp wi (9) W / (π / α) i where α is the regularization coefficient and W is the number of weight parameters in the ANN. The distribution in Eq.(9) is Gaussian with variance / α. When the magnitudes of weight parameters w s are large, p(w H) will be small. In this way, the prior distribution favors smaller values for weight parameters. Using the prior in Eq.(9), the posterior probability of the weight vector given the training data in Eq.(8) may be derived as [0] p( w D, H ) = exp( S( w) ) (0) ZS where α S(w ) = ECEE + w i () i and Z S ( α) = S( w) dw () According to Eq.(0), the weight vector that corresponds to the maximum of the posterior distribution w MP can be found by minimizing the negative logarithm of p(w D, H). This is equivalent to minimizing an error function in Eq.(), which consists of the cross-entropy error in Eq.(4) added to a weight penalty term called the weight decay regularizer. Evaluating the posterior probability in Eq.(0) is analytically intractable as it usually requires integration over a high dimensional weight space. One method of making these computations more tractable is to introduce approximations that simplify their computation. MacKay s [0] evidence framework and Buntine and Weigend [] approximate the posterior distribution, which is usually canonical, with a Gaussian distribution centered around one of its modes at w MP. Neal [] proposes a more general and exact approach where Markov Chain Monte Carlo methods are used to compute integrations over weight space. The 4

5 Neural Information Processing - Letters and Reviews Vol.0, No., January 006 method of MacKay [0] is used in this study because it is less computationally expensive than Neal s method [] and because of observations by Mackay [3] that the evidence approximation should produce superior results over Buntine and Weigend s method when estimating regularization coefficients. MacKay s evidence framework provides principled methods for model selection, prediction and estimating regularization coefficients without the need for a validation set of data. This allows more data to be used for training the ANN, which is useful when only limited amounts of data are available for training. The evidence framework specifies the following methods: ) Estimating Regularization Coefficients: The evidence of α is evaluated by integrating over network weight parameters as p ( D α, H ) = p( D w, H ) p( w α) dw (3) Using Eq.(3), MacKay [0] derives an expression for the most probable regularization coefficient α MP that maximises p(d α, H). This has the form of γ α = (4) where MP = W i i= λi + w i λ γ (5) α λ i s are eigenvectors of the Hessian matrix A or second derivative of the error function in Eq.() with respect to weight parameters. The Hessian is computed using the Pearlmutter method [4]. During training, α is first initialized to a small random value. Training takes place by adjusting weight parameters using the scaled conjugate gradient algorithm in order to minimize Eq.(). Once, a minimum is reached, α MP is determined using Eq.(4) keeping w constant. The above procedure is then repeated again until a self-consistent solution for both α and w is reached. ) Model Order Selection: The evidence framework compares ANN models of different complexity by evaluating the evidence of the model. For an ANN model H i, the evidence is given by: p( D Hi ) = p( D α, Hi) p(α Hi) dw (6) MacKay [] derives an expression for the logarithm of Eq.(6) as N 4π ( ) N! + ln γ W ln p( D Hi ) = S( w ) ln det( A) + lnα + ln (7) Studies have reported a correlation between the model evidence and generalization error of a neural network [0]. The expression in Eq.(7) allows the model with the highest evidence to be selected without having to separate training data into a training and validation set. 3) Moderated Outputs: ANNs estimate a mapping function by interpolating in a region of function space generated by a finite set of training data. The confidence of predictions within this space will be higher than predictions outside of this space. Moderated outputs were proposed by MacKay [0] to adjust the ANN output by an amount that is reflective of the uncertainty of the weight parameters. MacKay [0] derives an expression for the moderated output by assuming that the activation function of the output neuron, a, is a locally linear function of ANN weights. The expression can be written as: where and ( K( s ) P ( C x) = g ) (8) a MP ( / + π / 8) K ( s) = s (9) s ( x) g A g. (0) Here, a MP is the activation of the output neuron when w=w MP and g is the derivative of a with respect to w. = T 5

6 Detection of Epileptiform Activity in Human EEG Signals Figure. Main components of an automatic detection system. A sliding window segments the raw EEG data. Features are then extracted and presented to a classifier for classification. 4. SSW Detector Implementation The complete EEG SSW detection system is divided into three sequential components as shown in Fig.. The first stage is a data segmentation stage where the EEG is segmented into short, 30ms segments of data using a sliding window. Features are then extracted from the segments to provide a representation of the raw data more suitable for classification. A neural network classifier then classifies these features as containing epileptic activity or not. 4.. Feature Extraction Four different types of features were extracted from the EEG. These were based on methods used in previous studies. ) Fast Fourier Transform (FFT): The real and imaginary parts of the FFT of an EEG data segment were used by Jándo [4] as inputs to a neural network classifier and were reported to produce good detection performance. These features convey a spectral representation of an EEG data segment. A Gaussian window is applied to the data segment prior to the FFT in order to reduce spectral leakage. This window was chosen because of its good localization in time and frequency. ) Autoregressive Modeling: The Autoregressive Model has been used to extract features from the EEG in several EEG classification systems in the past [5]. This technique fits an EEG data segment to a parametric model in the form of a linear predictor. The parameters and prediction error of this model convey spectral information and are a low-dimensional representation of the raw data segment. A fourth order model was found to be sufficient to model the EEG. The Burg method was used to estimate the coefficients as it guarantees a stable model [6]. 3) Discrete Wavelet Transform (DWT): The above two feature extraction methods rely on an assumption of weak stationarity for the EEG data segment. However, in reality the EEG is a highly non-stationary signal. The DWT is suited to non-stationary signals and performs a multi-resolution analysis of a signal. Coefficients of the DWT are used as features that convey both temporal and spectral information. The method used by Kalayci and Özdamar [5] was adopted in this study. This method uses coefficients from the fourth level wavelet decomposition with a Daubechies-4 wavelet [7] as the mother wavelet 4) Raw EEG: Using the raw EEG data itself as input to a classifier has been shown to produce good results by Kalayci and Özdamar [8]. The benefit of using the raw data instead of extracted features is that the raw data will contain all information that is relevant for distinguishing between pattern classes, whereas extracted features may not represent all this information. 4.. Dimensionality Reduction The dimensionalities of the features were reduced using Principle Component Analysis (PCA) before presenting them to neural network classifiers for classification. Bishop [7] showed that a lower dimensional 6

7 Neural Information Processing - Letters and Reviews Vol.0, No., January 006 feature space will result in a less complex ANN that will require less training data for training. The amount of data required to train an ANN increases exponentially with dimensionality, the so called curse of dimensionality [7]. PCA or the Karhunen-Loéve transform is a linear unsupervised dimensionality reduction technique that aims to find a set of orthogonal vectors in the input data space that have the greatest contribution to the variance of the data. The eigenvectors and eigenvalues of the covariance matrix of the input features are calculated and only the largest eigenvalues are retained. Dimensionality reduction is achieved by projecting the input data onto these eigenvectors that correspond to these eigenvalues. PCA was found to reduce the dimensionality of raw EEG and FFT features by 75% and 80% while retaining 99% and 95% of the variance of the original data, respectively Input Standardization All input features are standardized to have a zero mean and unit standard deviation. Features extracted from the EEG differ significantly in magnitude. Larger inputs will be given higher importance by the ANN, which may not reflect their relative importance in making a classification decision. Standardization ensures that this does not occur and also assists with neural network training as smaller magnitude inputs are less likely to cause saturation of neural network activation functions Neural Network Validation Neural networks were implemented according to both the Maximum Likelihood and Bayesian Learning formulation. The generalization performance of each network was determined according to a four-fold cross validation strategy. The generalization error between ANN output and target output is quantified using a mean square error (MSE) criterion given by P EMSE = ( t p y p ) () P p= where P is the total number of test patterns, t p is the target output for input pattern p and y p is the actual ANN output when presented with pattern p. Classification performance is assessed in terms of Specificity, Sensitivity and Accuracy as defined below. ) Sensitivity: Percentage of input patterns correctly classified as containing epileptic activity. ) Specificity: Percentage of input patterns correctly classified as not containing epileptic activity 3) Accuracy: Percentage of correctly classified input patterns of either class Experimental Data The experimental data used to validate SSW detectors in this study was recorded from eleven patients with -bit accuracy at a sampling frequency of 00Hz using a Nihon-Kohden 00 digital EEG system. Nine of the patients EEG contained epileptiform activity and the remaining two exhibited normal brain activity. As the data is taken from several different patients, the ability of detectors to tolerate inter-patient variations in the EEG can be assessed. Data was marked as containing either epileptic activity, artifact activity or normal background activity by two EEG technologists, each having more than four years experience in the field. One thousand six hundred and seventy eight data segments and their associated markings were extracted from the dataset, of which half contained epileptiform activity and the remainder contained artifact or background activity. The markings were used to determine targets for neural network training and validation. 5. Results and Discussion Three different experiments were conducted in order to assess whether the methods described in this paper are useful in the context of epileptic SSW detection. In the first experiment, the mean square error on test data was determined for neural networks trained using both the Maximum Likelihood and Bayesian formulations. Note that this does not include the Bayesian moderated outputs as they are evaluated in a separate experiment. The results are shown in Table. For all four detectors, the Bayesian neural network consistently performs better regardless of type of input feature. This can probably be attributed to its more efficient use of training data during training. 7

8 Detection of Epileptiform Activity in Human EEG Signals Table. Mean Square Error on Test Dataset Type of Feature Network Configuration MSE Maximum Likelihood MSE Bayesian Raw FFT DWT AR Table. Classification Performance with Rejection Threshold of 0.9 Type of Feature Sensitivity Specificity Accuracy Not Classified Raw FFT DWT AR Table 3. Classification Performance with Rejection Threshold of 0.9 Moderated Outputs Type of Feature Sensitivity Specificity Accuracy Not Classified Raw FFT DWT AR The second experiment involves determining the classification performance of the Bayesian neural networks on test data. Classification decisions are made by applying a rejection threshold of 0.9 to the neural network outputs. This classification performance is computed using the test dataset and provides an estimate of a detector s true generalization performance. Results are shown in Table. The rejection thresholds produce very high values for sensitivity, specificity, and accuracy for all four detectors. As a result, the decisions made by the detectors appear very reliable. The detector with DWT input features appears to produce slightly superior results compared to other detectors. It is important to also study the amount of data that was not classified due to uncertainty of the neural network. The detector that uses parameters from the Autoregressive model as a feature is clearly inferior to all other detectors, because it does not classify 48.34% of features compared to between 4.97% and 7.% for the other detectors. Lastly, classification performance when using Bayesian Moderated Outputs was evaluated in order to assess whether their application could improve on the results in Table. These results are shown in Table 3. There is a negligible improvement in specificity, sensitivity and accuracy, and an increase in the number of unclassified features in comparison to the results in Table. We found that the moderated outputs required approximately ten times more computation time. This is mainly due to the computation of a Hessian matrix which is a computationally intensive task. As the Moderated Outputs do not improve classification results significantly and require additional computational overhead, we do not recommend their use for epileptic SSW detection. 6. Conclusion In this paper, we have implemented neural networks for epileptic SSW detection according to a Maximum Likelihood and Bayesian formulation. The latter was found to produce better trained networks and make better use of a finite set of training data than the former. Four different detector configurations were implemented with each detector using a different feature as input to a neural network. The DWT features appear to be a slightly better representation of the EEG compared to other features and also produce the least complex neural networks. Detectors that use AR features are significantly inferior to detectors that use other features, and are therefore not recommended for SSW detection. 8

9 Neural Information Processing - Letters and Reviews Vol.0, No., January 006 The use of rejection thresholds ensures that classification decisions are made with accuracies in excess of 90%. Doubtful input patterns are not classified and would have to be visually interpreted by a user of the detection tool. This is a desirable property for a detector as the perceived reliability of its classification decisions are high while still significantly reducing the amount of EEG data that needs to be visually screened. Bayesian moderated outputs do not improve on neural network predictions and add a significant computational overhead. Therefore, they are not recommended for use in SSW detection systems. The conclusion that can be drawn from this study is that neural networks are a powerful tool that is very well suited to EEG classification. It is important to interpret neural networks in a systematic manner within the wider framework of statistical pattern recognition. We have shown that reliable detection and a high degree of accuracy can be achieved when neural networks are implemented in this way. Acknowledgments The EEG data used in this study was collected by the Pattern Recognition Group of the Department of Electrical, Electronic and Computer Engineering in conjunction with the Department of Neurology at the University of Pretoria, South Africa. We gratefully acknowledge Professor E. C. Botha, Karl Geggus and Professor P. Bartel for making the data available for this work. References [l] J. R. Glover, P. Y. Ktonas, N. Raghaven, J. M., Urunela, S. V. Velamuri and E. L. Reilly, A multichannel signal processor for the detection of epileptogenic sharp transients in the EEG, IEEE Trans. Biomed. Engng, vol. BME-33(), pp. -8, 986 [] J. Gotman, P. Gloor and N. Schaul, Comparison of traditional reading of the EEG and automatic recognition of interictal epileptic activity, Elecroenceph. Clinical Neurophysiol., vol. 44, pp , 978. [3] J. Qian, J. S. Barlow and M. P. Beddoes, A simplified arithmetic detector for EEG sharp transients preliminary results, IEEE Trans.Biomed. Engng, vol. BME-35(), pp. -7, 988. [4] G. Jando, R. M. Siegel, Z. Horvath and G. Buzsaki, Pattern recognition of the electroencephalogram by artificial neural networks, Elecroenceph. Clinical Neurophysiol., vol. 86, pp , 993. [5] T. Kalayci and O. Ozdamar, Wavelet preprocessing for automated neural network detection of EEG spikes, IEEE Eng. Med. Biol. Mag., pp , Mar/Apr 995. [6] A. J. Gabor and M. Seyal, Automated interictal EEG spike detection using artificial neural networks, Elecroenceph. Clinical Neurophysiol., vol. 83, pp. 7-80, 99. [7] C. M. Bishop, Neural Networks for Pattern Recognition, Oxford University Press, Oxford, 995. [8] P. Sykacek, G. Dorffner, P. Rappelsberger and J. Zeitlhofer, Evaluating confidence measures in a neural network based sleep stager, Technical Report TR-97-, Austrian Research Institute for Artificial Intelligence (OFAI), 997. [9] M. F. Moller, A scaled conjugate gradient algorithm for fast supervised learning, Neural Networks, vol. 6, pp , 993. [0] D. J. C MacKay, Bayesian Methods for Adaptive Models, PhD thesis, California University of Technology, 99. [] W. L. Buntine and A. S. Weigend, Bayesian back-propagation, Complex Systems, vol. 5(6), pp , 99. [] R. M. Neal, Bayesian training of backpropagation networks by the Hybrid Monte Carlo method, Technical Report CRG-TR-9-, Connectionist Research Group, Department of Computer Science, University of Toronto, 99. [3] D. J. C. MacKay, Comparison of approximate methods for handling hyperparameters, Neural Computation, vol. (5), pp , 993. [4] B. A. Pearlmutter, Fast exact multiplication by the Hessian, Neural Computation, vol. 6(), pp , 994. [l5] A. Sharma and R. J. Roy, Design of a recognition system to predict movement during anaesthesia, IEEE Trans.Biomed. Engng, vol. 44, pp , 997 9

10 Detection of Epileptiform Activity in Human EEG Signals [l6] J. Pardey, S. Roberts and L. Tarassenko, A review of parametric modelling techniques for EEG analysis, Med. Eng. Physics, vol. 8, pp. -, 996 [7] I. Daubechies, Ten Lectures on Wavelets, Society for Industrial and Applied Mathematics, Philadelphia, 99. [8] T. Kalayci and O. Ozdamar, Detection of spikes with artificial neural networks using raw EEG, Comput. Biomedical Research, vol. 3, pp. -4, 998. Nadim Mohamed graduated from the University of the Witwatersrand with a BSc(Eng) and MSc(Eng) in Electrical Engineering in 003, and is currently a consultant at Accenture. His interests are in Biomedical Engineering, Bayesian learning and Pattern Recognition. David Rubin is a medical doctor and specialist in Nuclear Medicine. He received his medical degree, MBChB, from the University of Pretoria in 986. His other academic qualifications include Diploma in Anaesthetics and Fellow of the College of Nuclear Physicians, both from the South African College of Medicine, Master of Biomedical Engineering from the University of New South Wales, and the MMed degree from the University of the Witwatersrand, Johannesburg. He currently leads the Biomedical Engineering Research Group in the School of Electrical and Information Engineering at the University of the Witwatersrand, Johannesburg. His research interests include quantitative physiology, medical imaging, as well as identifying medical problems with potential engineering solutions, and working with engineers toward the realization of these solutions. Tshilidzi Marwala graduated with BS in Mechanical Engineering with a Magna Cum Laude from Case Western Reserve University in 995, a Masters of Engineering from University of Pretoria, and a PhD in Computational Intelligence from the University of Cambridge in 00. He was a postdoctoral research fellow at Imperial College of Science, Technology and Medicines. His interests are in System Identification and Pattern Recognition. 0

EEG signal classification using Bayes and Naïve Bayes Classifiers and extracted features of Continuous Wavelet Transform

EEG signal classification using Bayes and Naïve Bayes Classifiers and extracted features of Continuous Wavelet Transform EEG signal classification using Bayes and Naïve Bayes Classifiers and extracted features of Continuous Wavelet Transform Reza Yaghoobi Karimoi*, Mohammad Ali Khalilzadeh, Ali Akbar Hossinezadeh, Azra Yaghoobi

More information

CHAPTER 2 LITERATURE REVIEW

CHAPTER 2 LITERATURE REVIEW 9 CHAPTER 2 LITERATURE REVIEW In this chapter, a review of literature on Epileptic Seizure Detection, Wavelet Transform techniques, Principal Component Analysis, Artificial Neural Network, Radial Basis

More information

Epileptic seizure detection using EEG signals by means of stationary wavelet transforms

Epileptic seizure detection using EEG signals by means of stationary wavelet transforms I J C T A, 9(4), 2016, pp. 2065-2070 International Science Press Epileptic seizure detection using EEG signals by means of stationary wavelet transforms P. Grace Kanmani Prince 1, R. Rani Hemamalini 2,

More information

Performance Analysis of Epileptic EEG Expert System Using Scaled Conjugate Back Propagation Based ANN Classifier

Performance Analysis of Epileptic EEG Expert System Using Scaled Conjugate Back Propagation Based ANN Classifier Performance Analysis of Epileptic EEG Expert System Using Scaled Conjugate Back Propagation Based ANN Classifier Ashish Raj, Pankaj Gakhar, Meenu Kumari 3 Sweta Tripathi 4 ashishraj987@gmail.com,pankaj.gakhar@poornima.edu.in,meenu.kumari@poornima.edu.in

More information

Epileptic seizure detection using linear prediction filter

Epileptic seizure detection using linear prediction filter 11 th International conference on Sciences and Techniques of Automatic control & computer engineering December 19-1, 010, Monastir, Tunisia Epileptic seizure detection using linear prediction filter Introduction:

More information

ECG Beat Recognition using Principal Components Analysis and Artificial Neural Network

ECG Beat Recognition using Principal Components Analysis and Artificial Neural Network International Journal of Electronics Engineering, 3 (1), 2011, pp. 55 58 ECG Beat Recognition using Principal Components Analysis and Artificial Neural Network Amitabh Sharma 1, and Tanushree Sharma 2

More information

Classification of Epileptic EEG Using Wavelet Transform & Artificial Neural Network

Classification of Epileptic EEG Using Wavelet Transform & Artificial Neural Network Volume 4, No. 9, July-August 213 International Journal of Advanced Research in Computer Science RESEARCH PAPER Available Online at www.ijarcs.info ISSN No. 976-5697 Classification of Epileptic EEG Using

More information

arxiv: v1 [cs.lg] 4 Feb 2019

arxiv: v1 [cs.lg] 4 Feb 2019 Machine Learning for Seizure Type Classification: Setting the benchmark Subhrajit Roy [000 0002 6072 5500], Umar Asif [0000 0001 5209 7084], Jianbin Tang [0000 0001 5440 0796], and Stefan Harrer [0000

More information

NEURAL NETWORK CLASSIFICATION OF EEG SIGNAL FOR THE DETECTION OF SEIZURE

NEURAL NETWORK CLASSIFICATION OF EEG SIGNAL FOR THE DETECTION OF SEIZURE NEURAL NETWORK CLASSIFICATION OF EEG SIGNAL FOR THE DETECTION OF SEIZURE Shaguftha Yasmeen, M.Tech (DEC), Dept. of E&C, RIT, Bangalore, shagufthay@gmail.com Dr. Maya V Karki, Professor, Dept. of E&C, RIT,

More information

Epileptic Seizure Detection using Spike Information of Intrinsic Mode Functions with Neural Network

Epileptic Seizure Detection using Spike Information of Intrinsic Mode Functions with Neural Network Epileptic Seizure Detection using Spike Information of Intrinsic Mode Functions with Neural Network Gurwinder Singh Student (Mtech) Department of Computer Science Punjabi University Regional Centre for

More information

An Edge-Device for Accurate Seizure Detection in the IoT

An Edge-Device for Accurate Seizure Detection in the IoT An Edge-Device for Accurate Seizure Detection in the IoT M. A. Sayeed 1, S. P. Mohanty 2, E. Kougianos 3, and H. Zaveri 4 University of North Texas, Denton, TX, USA. 1,2,3 Yale University, New Haven, CT,

More information

Discrimination between ictal and seizure free EEG signals using empirical mode decomposition

Discrimination between ictal and seizure free EEG signals using empirical mode decomposition Discrimination between ictal and seizure free EEG signals using empirical mode decomposition by Ram Bilas Pachori in Accepted for publication in Research Letters in Signal Processing (Journal) Report No:

More information

EEG Signal Classification Using Wavelet Feature Extraction and Neural Networks

EEG Signal Classification Using Wavelet Feature Extraction and Neural Networks EEG Signal Classification Using Wavelet Feature Extraction and Neural Networks Pari Jahankhani, Vassilis Kodogiannis and Kenneth Revett AbstractDecision Support Systems have been utilised since 196, providing

More information

PCA Enhanced Kalman Filter for ECG Denoising

PCA Enhanced Kalman Filter for ECG Denoising IOSR Journal of Electronics & Communication Engineering (IOSR-JECE) ISSN(e) : 2278-1684 ISSN(p) : 2320-334X, PP 06-13 www.iosrjournals.org PCA Enhanced Kalman Filter for ECG Denoising Febina Ikbal 1, Prof.M.Mathurakani

More information

Epileptic Seizure Classification using Statistical Features of EEG Signal

Epileptic Seizure Classification using Statistical Features of EEG Signal International Conference on Electrical, Computer and Communication Engineering (ECCE), February 6-8,, Cox s Bazar, Bangladesh Epileptic Seizure Classification using Statistical Features of EEG Signal Md.

More information

Applying Data Mining for Epileptic Seizure Detection

Applying Data Mining for Epileptic Seizure Detection Applying Data Mining for Epileptic Seizure Detection Ying-Fang Lai 1 and Hsiu-Sen Chiang 2* 1 Department of Industrial Education, National Taiwan Normal University 162, Heping East Road Sec 1, Taipei,

More information

EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM

EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM EPILEPTIC SEIZURE DETECTION USING WAVELET TRANSFORM Sneha R. Rathod 1, Chaitra B. 2, Dr. H.P.Rajani 3, Dr. Rajashri khanai 4 1 MTech VLSI Design and Embedded systems,dept of ECE, KLE Dr.MSSCET, Belagavi,

More information

THE data used in this project is provided. SEIZURE forecasting systems hold promise. Seizure Prediction from Intracranial EEG Recordings

THE data used in this project is provided. SEIZURE forecasting systems hold promise. Seizure Prediction from Intracranial EEG Recordings 1 Seizure Prediction from Intracranial EEG Recordings Alex Fu, Spencer Gibbs, and Yuqi Liu 1 INTRODUCTION SEIZURE forecasting systems hold promise for improving the quality of life for patients with epilepsy.

More information

A Multistage, Multimethod Approach for Automatic Detection and Classification of Epileptiform EEG

A Multistage, Multimethod Approach for Automatic Detection and Classification of Epileptiform EEG IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 49, NO. 12, DECEMBER 2002 1557 A Multistage, Multimethod Approach for Automatic Detection and Classification of Epileptiform EEG He Sheng Liu 3, Tong Zhang,

More information

FREQUENCY DOMAIN BASED AUTOMATIC EKG ARTIFACT

FREQUENCY DOMAIN BASED AUTOMATIC EKG ARTIFACT FREQUENCY DOMAIN BASED AUTOMATIC EKG ARTIFACT REMOVAL FROM EEG DATA features FOR BRAIN such as entropy COMPUTER and kurtosis for INTERFACING artifact rejection. V. Viknesh B.E.,(M.E) - Lord Jeganath College

More information

Performance Analysis of Epileptic Seizure Detection Using DWT & ICA with Neural Networks

Performance Analysis of Epileptic Seizure Detection Using DWT & ICA with Neural Networks Performance Analysis of Epileptic Seizure Detection Using DWT & ICA with Neural Networks M. Stella Mercy Assistant Professor Kamaraj college of Engineering and Technology, Virudhunager, Tamilnadu, India.

More information

Minimum Feature Selection for Epileptic Seizure Classification using Wavelet-based Feature Extraction and a Fuzzy Neural Network

Minimum Feature Selection for Epileptic Seizure Classification using Wavelet-based Feature Extraction and a Fuzzy Neural Network Appl. Math. Inf. Sci. 8, No. 3, 129-1300 (201) 129 Applied Mathematics & Information Sciences An International Journal http://dx.doi.org/10.1278/amis/0803 Minimum Feature Selection for Epileptic Seizure

More information

DATA MANAGEMENT & TYPES OF ANALYSES OFTEN USED. Dennis L. Molfese University of Nebraska - Lincoln

DATA MANAGEMENT & TYPES OF ANALYSES OFTEN USED. Dennis L. Molfese University of Nebraska - Lincoln DATA MANAGEMENT & TYPES OF ANALYSES OFTEN USED Dennis L. Molfese University of Nebraska - Lincoln 1 DATA MANAGEMENT Backups Storage Identification Analyses 2 Data Analysis Pre-processing Statistical Analysis

More information

Automated Detection of Epileptic Seizures in the EEG

Automated Detection of Epileptic Seizures in the EEG 1 of 4 Automated Detection of Epileptic Seizures in the EEG Maarten-Jan Hoeve 1,, Richard D. Jones 1,3, Grant J. Carroll 4, Hansjerg Goelz 1 1 Department of Medical Physics & Bioengineering, Christchurch

More information

Robust system for patient specific classification of ECG signal using PCA and Neural Network

Robust system for patient specific classification of ECG signal using PCA and Neural Network International Research Journal of Engineering and Technology (IRJET) e-issn: 395-56 Volume: 4 Issue: 9 Sep -7 www.irjet.net p-issn: 395-7 Robust system for patient specific classification of using PCA

More information

CHAPTER IV PREPROCESSING & FEATURE EXTRACTION IN ECG SIGNALS

CHAPTER IV PREPROCESSING & FEATURE EXTRACTION IN ECG SIGNALS CHAPTER IV PREPROCESSING & FEATURE EXTRACTION IN ECG SIGNALS are The proposed ECG classification approach consists of three phases. They Preprocessing Feature Extraction and Selection Classification The

More information

A micropower support vector machine based seizure detection architecture for embedded medical devices

A micropower support vector machine based seizure detection architecture for embedded medical devices A micropower support vector machine based seizure detection architecture for embedded medical devices The MIT Faculty has made this article openly available. Please share how this access benefits you.

More information

Statistical analysis of epileptic activities based on histogram and wavelet-spectral entropy

Statistical analysis of epileptic activities based on histogram and wavelet-spectral entropy J. Biomedical Science and Engineering, 0, 4, 07-3 doi:0.436/jbise.0.4309 Published Online March 0 (http://www.scirp.org/journal/jbise/). Statistical analysis of epileptic activities based on histogram

More information

REVIEW ON ARRHYTHMIA DETECTION USING SIGNAL PROCESSING

REVIEW ON ARRHYTHMIA DETECTION USING SIGNAL PROCESSING REVIEW ON ARRHYTHMIA DETECTION USING SIGNAL PROCESSING Vishakha S. Naik Dessai Electronics and Telecommunication Engineering Department, Goa College of Engineering, (India) ABSTRACT An electrocardiogram

More information

Automatic Detection of Epileptic Seizures in EEG Using Machine Learning Methods

Automatic Detection of Epileptic Seizures in EEG Using Machine Learning Methods Automatic Detection of Epileptic Seizures in EEG Using Machine Learning Methods Ying-Fang Lai 1 and Hsiu-Sen Chiang 2* 1 Department of Industrial Education, National Taiwan Normal University 162, Heping

More information

Vital Responder: Real-time Health Monitoring of First- Responders

Vital Responder: Real-time Health Monitoring of First- Responders Vital Responder: Real-time Health Monitoring of First- Responders Ye Can 1,2 Advisors: Miguel Tavares Coimbra 2, Vijayakumar Bhagavatula 1 1 Department of Electrical & Computer Engineering, Carnegie Mellon

More information

Quick detection of QRS complexes and R-waves using a wavelet transform and K-means clustering

Quick detection of QRS complexes and R-waves using a wavelet transform and K-means clustering Bio-Medical Materials and Engineering 26 (2015) S1059 S1065 DOI 10.3233/BME-151402 IOS Press S1059 Quick detection of QRS complexes and R-waves using a wavelet transform and K-means clustering Yong Xia

More information

Computational Cognitive Neuroscience

Computational Cognitive Neuroscience Computational Cognitive Neuroscience Computational Cognitive Neuroscience Computational Cognitive Neuroscience *Computer vision, *Pattern recognition, *Classification, *Picking the relevant information

More information

Removal of Baseline wander and detection of QRS complex using wavelets

Removal of Baseline wander and detection of QRS complex using wavelets International Journal of Scientific & Engineering Research Volume 3, Issue 4, April-212 1 Removal of Baseline wander and detection of QRS complex using wavelets Nilesh Parihar, Dr. V. S. Chouhan Abstract

More information

Automatic Detection of Heart Disease Using Discreet Wavelet Transform and Artificial Neural Network

Automatic Detection of Heart Disease Using Discreet Wavelet Transform and Artificial Neural Network e-issn: 2349-9745 p-issn: 2393-8161 Scientific Journal Impact Factor (SJIF): 1.711 International Journal of Modern Trends in Engineering and Research www.ijmter.com Automatic Detection of Heart Disease

More information

A Learning Method of Directly Optimizing Classifier Performance at Local Operating Range

A Learning Method of Directly Optimizing Classifier Performance at Local Operating Range A Learning Method of Directly Optimizing Classifier Performance at Local Operating Range Lae-Jeong Park and Jung-Ho Moon Department of Electrical Engineering, Kangnung National University Kangnung, Gangwon-Do,

More information

Selection of Feature for Epilepsy Seizer Detection Using EEG

Selection of Feature for Epilepsy Seizer Detection Using EEG International Journal of Neurosurgery 2018; 2(1): 1-7 http://www.sciencepublishinggroup.com/j/ijn doi: 10.11648/j.ijn.20180201.11 Selection of Feature for Epilepsy Seizer Detection Using EEG Manisha Chandani

More information

Neural Network based Heart Arrhythmia Detection and Classification from ECG Signal

Neural Network based Heart Arrhythmia Detection and Classification from ECG Signal Neural Network based Heart Arrhythmia Detection and Classification from ECG Signal 1 M. S. Aware, 2 V. V. Shete *Dept. of Electronics and Telecommunication, *MIT College Of Engineering, Pune Email: 1 mrunal_swapnil@yahoo.com,

More information

Application of Artificial Neural Networks in Classification of Autism Diagnosis Based on Gene Expression Signatures

Application of Artificial Neural Networks in Classification of Autism Diagnosis Based on Gene Expression Signatures Application of Artificial Neural Networks in Classification of Autism Diagnosis Based on Gene Expression Signatures 1 2 3 4 5 Kathleen T Quach Department of Neuroscience University of California, San Diego

More information

Automatic Definition of Planning Target Volume in Computer-Assisted Radiotherapy

Automatic Definition of Planning Target Volume in Computer-Assisted Radiotherapy Automatic Definition of Planning Target Volume in Computer-Assisted Radiotherapy Angelo Zizzari Department of Cybernetics, School of Systems Engineering The University of Reading, Whiteknights, PO Box

More information

Wavelet Neural Network for Classification of Bundle Branch Blocks

Wavelet Neural Network for Classification of Bundle Branch Blocks , July 6-8, 2011, London, U.K. Wavelet Neural Network for Classification of Bundle Branch Blocks Rahime Ceylan, Yüksel Özbay Abstract Bundle branch blocks are very important for the heart treatment immediately.

More information

Empirical Mode Decomposition based Feature Extraction Method for the Classification of EEG Signal

Empirical Mode Decomposition based Feature Extraction Method for the Classification of EEG Signal Empirical Mode Decomposition based Feature Extraction Method for the Classification of EEG Signal Anant kulkarni MTech Communication Engineering Vellore Institute of Technology Chennai, India anant8778@gmail.com

More information

Correlation analysis of seizure detection features

Correlation analysis of seizure detection features Correlation analysis of seizure detection features # L. Kuhlmann, M. J. Cook 2,3, K. Fuller 2, D. B. Grayden,3, A. N. Burkitt,3, I.M.Y. Mareels Department of Electrical and Electronic Engineering, University

More information

Epileptic Seizure Classification of EEG Image Using ANN

Epileptic Seizure Classification of EEG Image Using ANN Epileptic Seizure Classification of EEG Image Using ANN Prof. (Dr.) M.K. Bhaskar Professor, Electrical Engg. Department M.B.M. Engg. College, Jodhpur, Raj, India Prof. Surendra. Bohra Professor, ECE Department

More information

LIE DETECTION SYSTEM USING INPUT VOICE SIGNAL K.Meena 1, K.Veena 2 (Corresponding Author: K.Veena) 1 Associate Professor, 2 Research Scholar,

LIE DETECTION SYSTEM USING INPUT VOICE SIGNAL K.Meena 1, K.Veena 2 (Corresponding Author: K.Veena) 1 Associate Professor, 2 Research Scholar, International Journal of Pure and Applied Mathematics Volume 117 No. 8 2017, 121-125 ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version) url: http://www.ijpam.eu doi: 10.12732/ijpam.v117i8.25

More information

NEURAL NETWORK CLASSIFICATION OF EEG SIGNALS BY USING AR WITH MLE PREPROCESSING FOR EPILEPTIC SEIZURE DETECTION

NEURAL NETWORK CLASSIFICATION OF EEG SIGNALS BY USING AR WITH MLE PREPROCESSING FOR EPILEPTIC SEIZURE DETECTION Mathematical and Computational Applications, Vol. 10, No. 1, pp. 57-70, 2005. Association for Scientific Research NEURAL NETWORK CLASSIFICATION OF EEG SIGNALS BY USING AR WITH MLE PREPROCESSING FOR EPILEPTIC

More information

Seizure Prediction Through Clustering and Temporal Analysis of Micro Electrocorticographic Data

Seizure Prediction Through Clustering and Temporal Analysis of Micro Electrocorticographic Data Seizure Prediction Through Clustering and Temporal Analysis of Micro Electrocorticographic Data Yilin Song 1, Jonathan Viventi 2, and Yao Wang 1 1 Department of Electrical and Computer Engineering, New

More information

A HMM-based Pre-training Approach for Sequential Data

A HMM-based Pre-training Approach for Sequential Data A HMM-based Pre-training Approach for Sequential Data Luca Pasa 1, Alberto Testolin 2, Alessandro Sperduti 1 1- Department of Mathematics 2- Department of Developmental Psychology and Socialisation University

More information

Chapter 1. Introduction

Chapter 1. Introduction Chapter 1 Introduction 1.1 Motivation and Goals The increasing availability and decreasing cost of high-throughput (HT) technologies coupled with the availability of computational tools and data form a

More information

CHAPTER 5 WAVELET BASED DETECTION OF VENTRICULAR ARRHYTHMIAS WITH NEURAL NETWORK CLASSIFIER

CHAPTER 5 WAVELET BASED DETECTION OF VENTRICULAR ARRHYTHMIAS WITH NEURAL NETWORK CLASSIFIER 57 CHAPTER 5 WAVELET BASED DETECTION OF VENTRICULAR ARRHYTHMIAS WITH NEURAL NETWORK CLASSIFIER 5.1 INTRODUCTION The cardiac disorders which are life threatening are the ventricular arrhythmias such as

More information

Fuzzy Based Early Detection of Myocardial Ischemia Using Wavelets

Fuzzy Based Early Detection of Myocardial Ischemia Using Wavelets Fuzzy Based Early Detection of Myocardial Ischemia Using Wavelets Jyoti Arya 1, Bhumika Gupta 2 P.G. Student, Department of Computer Science, GB Pant Engineering College, Ghurdauri, Pauri, India 1 Assistant

More information

Enhanced Detection of Lung Cancer using Hybrid Method of Image Segmentation

Enhanced Detection of Lung Cancer using Hybrid Method of Image Segmentation Enhanced Detection of Lung Cancer using Hybrid Method of Image Segmentation L Uma Maheshwari Department of ECE, Stanley College of Engineering and Technology for Women, Hyderabad - 500001, India. Udayini

More information

Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons

Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons Development of 2-Channel Eeg Device And Analysis Of Brain Wave For Depressed Persons P.Amsaleka*, Dr.S.Mythili ** * PG Scholar, Applied Electronics, Department of Electronics and Communication, PSNA College

More information

Neurons and neural networks II. Hopfield network

Neurons and neural networks II. Hopfield network Neurons and neural networks II. Hopfield network 1 Perceptron recap key ingredient: adaptivity of the system unsupervised vs supervised learning architecture for discrimination: single neuron perceptron

More information

Analysis of Speech Recognition Techniques for use in a Non-Speech Sound Recognition System

Analysis of Speech Recognition Techniques for use in a Non-Speech Sound Recognition System Analysis of Recognition Techniques for use in a Sound Recognition System Michael Cowling, Member, IEEE and Renate Sitte, Member, IEEE Griffith University Faculty of Engineering & Information Technology

More information

Automated Prediction of Thyroid Disease using ANN

Automated Prediction of Thyroid Disease using ANN Automated Prediction of Thyroid Disease using ANN Vikram V Hegde 1, Deepamala N 2 P.G. Student, Department of Computer Science and Engineering, RV College of, Bangalore, Karnataka, India 1 Assistant Professor,

More information

CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL

CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL 116 CHAPTER 6 INTERFERENCE CANCELLATION IN EEG SIGNAL 6.1 INTRODUCTION Electrical impulses generated by nerve firings in the brain pass through the head and represent the electroencephalogram (EEG). Electrical

More information

Reading Assignments: Lecture 18: Visual Pre-Processing. Chapters TMB Brain Theory and Artificial Intelligence

Reading Assignments: Lecture 18: Visual Pre-Processing. Chapters TMB Brain Theory and Artificial Intelligence Brain Theory and Artificial Intelligence Lecture 18: Visual Pre-Processing. Reading Assignments: Chapters TMB2 3.3. 1 Low-Level Processing Remember: Vision as a change in representation. At the low-level,

More information

Cognitive Neuroscience History of Neural Networks in Artificial Intelligence The concept of neural network in artificial intelligence

Cognitive Neuroscience History of Neural Networks in Artificial Intelligence The concept of neural network in artificial intelligence Cognitive Neuroscience History of Neural Networks in Artificial Intelligence The concept of neural network in artificial intelligence To understand the network paradigm also requires examining the history

More information

PMR5406 Redes Neurais e Lógica Fuzzy. Aula 5 Alguns Exemplos

PMR5406 Redes Neurais e Lógica Fuzzy. Aula 5 Alguns Exemplos PMR5406 Redes Neurais e Lógica Fuzzy Aula 5 Alguns Exemplos APPLICATIONS Two examples of real life applications of neural networks for pattern classification: RBF networks for face recognition FF networks

More information

A Kalman filter based methodology for EEG spike enhancement

A Kalman filter based methodology for EEG spike enhancement computer methods and programs in biomedicine 85 (2007) 101 108 journal homepage: www.intl.elsevierhealth.com/journals/cmpb A Kalman filter based methodology for EEG spike enhancement V.P. Oikonomou a,

More information

EEG SPIKE CLASSIFICATION WITH TEMPLATE MATCHING ALGORITHM. Çamlık Caddesi No:44 Sarnıç Beldesi İZMİR 2 Elektrik ve Elektronik Müh.

EEG SPIKE CLASSIFICATION WITH TEMPLATE MATCHING ALGORITHM. Çamlık Caddesi No:44 Sarnıç Beldesi İZMİR 2 Elektrik ve Elektronik Müh. EEG SPIKE CLASSIFICATION WITH TEMPLATE MATCHING ALGORITHM Selim BÖLGEN 1 Gülden KÖKTÜRK 2 1 Pagetel Sistem Müh. San. Tic. Ltd. Şti. Çamlık Caddesi No:44 Sarnıç Beldesi İZMİR 2 Elektrik ve Elektronik Müh.

More information

Classification of ECG Data for Predictive Analysis to Assist in Medical Decisions.

Classification of ECG Data for Predictive Analysis to Assist in Medical Decisions. 48 IJCSNS International Journal of Computer Science and Network Security, VOL.15 No.10, October 2015 Classification of ECG Data for Predictive Analysis to Assist in Medical Decisions. A. R. Chitupe S.

More information

Information Processing During Transient Responses in the Crayfish Visual System

Information Processing During Transient Responses in the Crayfish Visual System Information Processing During Transient Responses in the Crayfish Visual System Christopher J. Rozell, Don. H. Johnson and Raymon M. Glantz Department of Electrical & Computer Engineering Department of

More information

Bayesian and Frequentist Approaches

Bayesian and Frequentist Approaches Bayesian and Frequentist Approaches G. Jogesh Babu Penn State University http://sites.stat.psu.edu/ babu http://astrostatistics.psu.edu All models are wrong But some are useful George E. P. Box (son-in-law

More information

An ECG Beat Classification Using Adaptive Neuro- Fuzzy Inference System

An ECG Beat Classification Using Adaptive Neuro- Fuzzy Inference System An ECG Beat Classification Using Adaptive Neuro- Fuzzy Inference System Pramod R. Bokde Department of Electronics Engineering, Priyadarshini Bhagwati College of Engineering, Nagpur, India Abstract Electrocardiography

More information

EEG Interictal Spike Detection Using Artificial Neural Networks

EEG Interictal Spike Detection Using Artificial Neural Networks Virginia Commonwealth University VCU Scholars Compass Theses and Dissertations Graduate School 2016 EEG Interictal Spike Detection Using Artificial Neural Networks Howard J. Carey III Virginia Commonwealth

More information

Analysis of EEG Signal for the Detection of Brain Abnormalities

Analysis of EEG Signal for the Detection of Brain Abnormalities Analysis of EEG Signal for the Detection of Brain Abnormalities M.Kalaivani PG Scholar Department of Computer Science and Engineering PG National Engineering College Kovilpatti, Tamilnadu V.Kalaivani,

More information

INCLUSION OF ECG AND EEG ANALYSIS IN NEURAL NETWORK MODELS

INCLUSION OF ECG AND EEG ANALYSIS IN NEURAL NETWORK MODELS INCLUSION OF ECG AND EEG ANALYSIS IN NEURAL NETWORK MODELS Maurice E. Cohen 1,2, Donna L. Hudson 1 1 California State University, Fresno, CA 93740, USA 2 University of California, San Francisco, Fresno,

More information

CARDIAC ARRYTHMIA CLASSIFICATION BY NEURONAL NETWORKS (MLP)

CARDIAC ARRYTHMIA CLASSIFICATION BY NEURONAL NETWORKS (MLP) CARDIAC ARRYTHMIA CLASSIFICATION BY NEURONAL NETWORKS (MLP) Bochra TRIQUI, Abdelkader BENYETTOU Center for Artificial Intelligent USTO-MB University Algeria triqui_bouchra@yahoo.fr a_benyettou@yahoo.fr

More information

Epilepsy Disorder Detection from EEG Signal

Epilepsy Disorder Detection from EEG Signal Int.J. of Intelligent Computing and Applied Sciences 41 Epilepsy Disorder Detection from EEG Signal Pradipta Kumar Das* Department of Computer Science and Engineering Dhaneswar Rath Institute of Engineering

More information

Intelligent Edge Detector Based on Multiple Edge Maps. M. Qasim, W.L. Woon, Z. Aung. Technical Report DNA # May 2012

Intelligent Edge Detector Based on Multiple Edge Maps. M. Qasim, W.L. Woon, Z. Aung. Technical Report DNA # May 2012 Intelligent Edge Detector Based on Multiple Edge Maps M. Qasim, W.L. Woon, Z. Aung Technical Report DNA #2012-10 May 2012 Data & Network Analytics Research Group (DNA) Computing and Information Science

More information

ERA: Architectures for Inference

ERA: Architectures for Inference ERA: Architectures for Inference Dan Hammerstrom Electrical And Computer Engineering 7/28/09 1 Intelligent Computing In spite of the transistor bounty of Moore s law, there is a large class of problems

More information

Bayesian Models for Combining Data Across Subjects and Studies in Predictive fmri Data Analysis

Bayesian Models for Combining Data Across Subjects and Studies in Predictive fmri Data Analysis Bayesian Models for Combining Data Across Subjects and Studies in Predictive fmri Data Analysis Thesis Proposal Indrayana Rustandi April 3, 2007 Outline Motivation and Thesis Preliminary results: Hierarchical

More information

Genetic Algorithm based Feature Extraction for ECG Signal Classification using Neural Network

Genetic Algorithm based Feature Extraction for ECG Signal Classification using Neural Network Genetic Algorithm based Feature Extraction for ECG Signal Classification using Neural Network 1 R. Sathya, 2 K. Akilandeswari 1,2 Research Scholar 1 Department of Computer Science 1 Govt. Arts College,

More information

Linear and Nonlinear Optimization

Linear and Nonlinear Optimization Linear and Nonlinear Optimization SECOND EDITION Igor Griva Stephen G. Nash Ariela Sofer George Mason University Fairfax, Virginia Society for Industrial and Applied Mathematics Philadelphia Contents Preface

More information

Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter Detection

Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter Detection Proceedings of the 2012 International Conference on Industrial Engineering and Operations Management Istanbul, Turkey, July 3 6, 2012 Assessment of Reliability of Hamilton-Tompkins Algorithm to ECG Parameter

More information

UNIVERSITY of PENNSYLVANIA CIS 520: Machine Learning Final, Fall 2014

UNIVERSITY of PENNSYLVANIA CIS 520: Machine Learning Final, Fall 2014 UNIVERSITY of PENNSYLVANIA CIS 520: Machine Learning Final, Fall 2014 Exam policy: This exam allows two one-page, two-sided cheat sheets (i.e. 4 sides); No other materials. Time: 2 hours. Be sure to write

More information

10CS664: PATTERN RECOGNITION QUESTION BANK

10CS664: PATTERN RECOGNITION QUESTION BANK 10CS664: PATTERN RECOGNITION QUESTION BANK Assignments would be handed out in class as well as posted on the class blog for the course. Please solve the problems in the exercises of the prescribed text

More information

Learning in neural networks

Learning in neural networks http://ccnl.psy.unipd.it Learning in neural networks Marco Zorzi University of Padova M. Zorzi - European Diploma in Cognitive and Brain Sciences, Cognitive modeling", HWK 19-24/3/2006 1 Connectionist

More information

Performance of Gaussian Mixture Models as a Classifier for Pathological Voice

Performance of Gaussian Mixture Models as a Classifier for Pathological Voice PAGE 65 Performance of Gaussian Mixture Models as a Classifier for Pathological Voice Jianglin Wang, Cheolwoo Jo SASPL, School of Mechatronics Changwon ational University Changwon, Gyeongnam 64-773, Republic

More information

Introduction to Machine Learning. Katherine Heller Deep Learning Summer School 2018

Introduction to Machine Learning. Katherine Heller Deep Learning Summer School 2018 Introduction to Machine Learning Katherine Heller Deep Learning Summer School 2018 Outline Kinds of machine learning Linear regression Regularization Bayesian methods Logistic Regression Why we do this

More information

Emotion Detection Using Physiological Signals. M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis

Emotion Detection Using Physiological Signals. M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis Emotion Detection Using Physiological Signals M.A.Sc. Thesis Proposal Haiyan Xu Supervisor: Prof. K.N. Plataniotis May 10 th, 2011 Outline Emotion Detection Overview EEG for Emotion Detection Previous

More information

EEG Signal Classification using Fusion of DWT, SWT and PCA Features

EEG Signal Classification using Fusion of DWT, SWT and PCA Features EEG Signal Classification using Fusion of DWT, SWT and PCA Features Rohini Darade 1, Prof. S. R. Baji 2 1 E&TC Dept, LGNSCOE, Nashik 2 E&TC Dept, LGNSCOE, Nashik Abstract Human brain is a diverse creature,

More information

PARAFAC: a powerful tool in EEG monitoring

PARAFAC: a powerful tool in EEG monitoring Katholieke Universiteit Leuven K.U.Leuven PARAFAC: a powerful tool in EEG monitoring Sabine Van Huffel Dept. Electrical Engineering ESAT-SCD SCD Katholieke Universiteit Leuven, Belgium 1 Contents Overview

More information

Australian Journal of Basic and Applied Sciences

Australian Journal of Basic and Applied Sciences Australian Journal of Basic and Applied Sciences, 7(12) October 2013, Pages: 174-179 AENSI Journals Australian Journal of Basic and Applied Sciences Journal home page: www.ajbasweb.com A Probabilistic

More information

Fuzzy Techniques for Classification of Epilepsy risk level in Diabetic Patients Using Cerebral Blood Flow and Aggregation Operators

Fuzzy Techniques for Classification of Epilepsy risk level in Diabetic Patients Using Cerebral Blood Flow and Aggregation Operators Fuzzy Techniques for Classification of Epilepsy risk level in Diabetic Patients Using Cerebral Blood Flow and Aggregation Operators R.Harikumar, Dr. (Mrs).R.Sukanesh Research Scholar Assistant Professor

More information

Implementation of Probabilistic Neural Network using Approximate Entropy to Detect Epileptic Seizures

Implementation of Probabilistic Neural Network using Approximate Entropy to Detect Epileptic Seizures Implementation of Probabilistic Neural Network using Approximate Entropy to Detect Epileptic Seizures Sachee 1, Roohi Sille 2, Garima Sharma 3 & N. Pradhan 4 1,2&3 Dept. of Biomedical Engineering, Bundelkhand

More information

Feasibility Study of the Time-variant Functional Connectivity Pattern during an Epileptic Seizure

Feasibility Study of the Time-variant Functional Connectivity Pattern during an Epileptic Seizure International Journal of Bioelectromagnetism Vol. 11, No. 4, pp.170-174, 009 www.ijbem.org Feasibility Study of the Time-variant Functional Connectivity Pattern during an Epileptic Seizure Pieter van Mierlo

More information

A MULTI-STAGE NEURAL NETWORK CLASSIFIER FOR ECG EVENTS

A MULTI-STAGE NEURAL NETWORK CLASSIFIER FOR ECG EVENTS A MULTI-STAGE NEURAL NETWORK CLASSIFIER FOR ECG EVENTS H. Gholam Hosseini 1, K. J. Reynolds 2, D. Powers 2 1 Department of Electrotechnology, Auckland University of Technology, Auckland, New Zealand 2

More information

Research Article Detection of Epileptic Seizure Event and Onset Using EEG

Research Article Detection of Epileptic Seizure Event and Onset Using EEG BioMed Research International, Article ID 45573, 7 pages http://dx.doi.org/1.1155//45573 Research Article Detection of Epileptic Seizure Event and Onset Using EEG abeel Ahammad, Thasneem Fathima, and Paul

More information

Speech recognition in noisy environments: A survey

Speech recognition in noisy environments: A survey T-61.182 Robustness in Language and Speech Processing Speech recognition in noisy environments: A survey Yifan Gong presented by Tapani Raiko Feb 20, 2003 About the Paper Article published in Speech Communication

More information

Question 1 Multiple Choice (8 marks)

Question 1 Multiple Choice (8 marks) Philadelphia University Student Name: Faculty of Engineering Student Number: Dept. of Computer Engineering First Exam, First Semester: 2015/2016 Course Title: Neural Networks and Fuzzy Logic Date: 19/11/2015

More information

Processed by HBI: Russia/Switzerland/USA

Processed by HBI: Russia/Switzerland/USA 1 CONTENTS I Personal and clinical data II Conclusion. III Recommendations for therapy IV Report. 1. Procedures of EEG recording and analysis 2. Search for paroxysms 3. Eyes Open background EEG rhythms

More information

Frequency Tracking: LMS and RLS Applied to Speech Formant Estimation

Frequency Tracking: LMS and RLS Applied to Speech Formant Estimation Aldebaro Klautau - http://speech.ucsd.edu/aldebaro - 2/3/. Page. Frequency Tracking: LMS and RLS Applied to Speech Formant Estimation ) Introduction Several speech processing algorithms assume the signal

More information

Detection of pulmonary abnormalities using Multi scale products and ARMA modelling

Detection of pulmonary abnormalities using Multi scale products and ARMA modelling Volume 119 No. 15 2018, 2177-2181 ISSN: 1314-3395 (on-line version) url: http://www.acadpubl.eu/hub/ http://www.acadpubl.eu/hub/ Detection of pulmonary abnormalities using Multi scale products and ARMA

More information

CHAPTER 4 CLASSIFICATION OF HEART MURMURS USING WAVELETS AND NEURAL NETWORKS

CHAPTER 4 CLASSIFICATION OF HEART MURMURS USING WAVELETS AND NEURAL NETWORKS 52 CHAPTER 4 CLASSIFICATION OF HEART MURMURS USING WAVELETS AND NEURAL NETWORKS 4.1 INTRODUCTION Heart auscultation is the process of interpreting sounds produced by the turbulent flow of blood into and

More information

Classification of EEG signals in an Object Recognition task

Classification of EEG signals in an Object Recognition task Classification of EEG signals in an Object Recognition task Iacob D. Rus, Paul Marc, Mihaela Dinsoreanu, Rodica Potolea Technical University of Cluj-Napoca Cluj-Napoca, Romania 1 rus_iacob23@yahoo.com,

More information

International Journal of Research in Science and Technology. (IJRST) 2018, Vol. No. 8, Issue No. IV, Oct-Dec e-issn: , p-issn: X

International Journal of Research in Science and Technology. (IJRST) 2018, Vol. No. 8, Issue No. IV, Oct-Dec e-issn: , p-issn: X CLOUD FILE SHARING AND DATA SECURITY THREATS EXPLORING THE EMPLOYABILITY OF GRAPH-BASED UNSUPERVISED LEARNING IN DETECTING AND SAFEGUARDING CLOUD FILES Harshit Yadav Student, Bal Bharati Public School,

More information

Primary Tumor detection with EEG Signals using Wavelet Transform and Neural Network

Primary Tumor detection with EEG Signals using Wavelet Transform and Neural Network Primary Tumor detection with EEG Signals using Wavelet Transform and Neural Network Mr. Ankush Surkar Prof. Nitin Ambatkar M. Tech Student, Department of ETC Ass. Prof., Department of ETC, Priyadarshni

More information