Ertugrul Yazgan

Detection of ECG waveforms by neural networks.

In this study, ECG waveform detection was performed by using artificial neural networks (ANNs). Initially, the R peak of the QRS complex is detected, and then feature vectors are formed by using the amplitudes of the significant frequency components of the DFT spectrum. Grow and Learn (GAL) and Kohonen networks are comparatively investigated to detect four different ECG waveforms. The comparative performance results of GAL and Kohonen networks are reported.

Compensatory fuzzy neural networks-based intelligent detection of abnormal neonatal cerebral Doppler ultrasound waveforms

Compensatory fuzzy neural networks (CFNN) without normalization, which can be trained with a backpropagation learning algorithm, are proposed as a pattern recognition technique for the intelligent detection of Doppler ultrasound waveforms of abnormal neonatal cerebral hemodynamics. Doppler ultrasound signals were recorded from the anterior cerebral arteries of 40 normal full-term babies and 14 mature babies with intracranial pathology. The features of normal and abnormal groups as inputs to the pattern recognition algorithms were extracted from the maximum-velocity waveforms by using principal component analysis. The proposed technique is compared with the CFNN with normalization and other pattern recognition techniques applied to Doppler ultrasound signals from various arteries. The results show that the proposed method is superior to the other techniques, and can be a powerful way to analyze Doppler ultrasound signals from various arteries.

A noise reduction algorithm in ECG signals using wavelet transform

A noise reduction algorithm based on the multiresolution orthogonal wavelet transform for ECG signals is discussed. Results are presented for different analysis and synthesis filter coefficients proposed by Vetterli-Herley and Daubechies for two different noise characteristics. The first one is artificially obtained line interference sinusoidal noise and the second one is uniformly distributed random noise or white noise. The results show that the former noise is completely eliminated front the signal by using both of the filter coefficients. However. The latter noise did not give the same performance as that of the former one.

Classification of ECG waveforms using a novel neural network

It is important to detect and display waveforms on the ECG recordings fast and automatically. In this study, an artificial neural network trained by genetic algorithms (NeTGA) is proposed for ECG waveform detection. After the R peak of the QRS complex is detected, a window containing an ECG period is formed around the R peak. The significant frequency components of the signal in the window are used to form the feature vectors. NeTGA, grow and learn (GAL), multi-layer perceptron (MLP), and Kohonen networks are comparatively investigated to detect seven different ECG waveforms. It is observed that the proposed network results in better classification performance with less number of nodes.

Classification of MR and CT images using genetic algorithms

A modified restricted Coulomb energy (MoRCE) network trained by the genetic algorithm is presented. Each neuron of the network forms a closed region in the input space. The closed regions which are formed by the neurons overlap each other, like STAR. Genetic algorithms are used to improve the classification performances of the magnetic resonance (MR) and computer tomography (CT) images with minimized number of neurons. MoRCE is examined comparatively with multilayer perceptron (MLP), and restricted Coulomb energy (RCE). It is observed that MoRCE gives the best classification performance with less number of neurons after a short training time.

ECG waveform classification using the neural network and wavelet transform

Two feature extraction methods: Fourier analysis and wavelet analysis for ECG waveform classification are comparatively investigated. Ten different ECG waveforms from MIT/BIH database are classified using a neural network trained by genetic algorithms (NeTGA). One set of feature vectors is formed by using DFT coefficients, and the second set is formed by using wavelet transform (WT) coefficients and their autocorrelation values. Elements of the feature vectors are searched by using dynamic programming (DP) according to the divergence values. Wavelet feature set is found to result in better classification accuracy with less number of nodes. It is observed that with the feature set formed by wavelet analysis, NeTGA gives 99.4% classification performance with 26 nodes after a short training time.

Classification of ECG waveforms by using genetic algorithms

In this study, a restricted coulomb energy network trained by genetic algorithms (GARCE) is proposed for ECG (electrocardiogram) waveform detection. After the R peak of the QRS complex is detected, a window containing an ECG period is formed around the R peak. The significant frequency components of the discrete Fourier transform of the signal in this window are used to form the feature vectors. Restricted Coulomb energy (RCE), multilayer perceptron (MLP) and GARCE networks are comparatively examined to detect 7 different ECG waveforms. The comparative performance results of these networks indicate that the GARCE network results in faster learning and better classification performance with less number of nodes.

Abnormal tissue detection in computer tomography images using artificial neural networks

Artificial neural network solutions have been applied to numerous image applications in the recent past. This paper describes a method to detect abnormal tissues in computer tomography (CT) head images by using an artificial neural network. We use a three-layer perceptron: An input layer with 882 nodes, one hidden layer with 30 nodes and an output layer with 3 nodes. The output nodes represent the bone, soft and abnormal tissue. The input layer of network does not directly receive the data from the original image. Two processes are applied to the pixels in subimage which has 7×7 pixels, and the results are fed to the input layer of the network. These processes provide necessary information for the tissue identification.

Detection of ECG waveforms by using artificial neural networks

The ECG has considerable diagnostic significance in medicine. It is important to detect and display waveforms on the ECG recordings fast and automatically. In this study, waveform detection is performed by using artificial neural networks (ANNs). After the detection of the R peak of the QRS complex, feature vectors are formed by using the amplitudes of the significant frequency components of the DFT frequency spectrum. Grow and Learn (GAL) and Kohonen networks are comparatively examined to detect 4 different ECG waveforms. The comparative performance results of GAL, and Kohonen networks indicate that the GAL network results in faster learning and better classification performance with less number of nodes.

MR image classification by the neural network and the genetic algorithms

A novel neural network trained by the genetic algorithms (GAs) is presented. Each neuron of the network forms a closed region in an input space. The locations of the centers of the closed regions (CR) are optimized in order to minimize the number of the neurons used and to improve the classification performance. After the network is trained by the set which is formed by the supervisor, it is used to classify a magnetic resonance (MR) image with a tumor.