Implementing wavelet transform/mixture of experts network for analysis of electrocardiogram beats

This paper illustrates the use of mixture of experts (ME) network structure to guide model selection for classification of electroencephalogram (EEG) signals. Expectation-maximization (EM) algorithm was used for training the ME so that the learning process is decoupled in a manner that fits well with the modular structure. The EEG signals were decomposed into time-frequency representations using discrete wavelet transform and statistical features were calculated to depict their distribution. The ME network structure was implemented for classification of the EEG signals using the statistical features as inputs. Three types of EEG signals (EEG signals recorded from healthy volunteers with eyes open, epilepsy patients in the epileptogenic zone during a seizure-free interval, and epilepsy patients during epileptic seizures) were classified with the accuracy of 93.17% by the ME network structure. The ME network structure achieved accuracy rates which were higher than that of the stand-alone neural network models.

EEG signal classification using wavelet feature extraction and a mixture of expert model

In this paper, an electrocardiogram (ECG) beat classification system based on wavelet transformation and probabilistic neural network (PNN) is proposed to discriminate six ECG beat types. The effects of two wavelet decomposition structures, the two-stage two-band and the two-stage full binary decomposition structures, in the recognition of ECG beat types are studied. The ECG beat signals are first decomposed into components in different subbands using discrete wavelet transformation. Three statistical features of each decomposed subband signals as well as the AC power and instantaneous RR interval of the original signal are exploited to characterize the ECG signals. A PNN then follows to classify the feature vectors. The result shows that features extracted from the decomposed signals based on the two-stage two-band structure outperform the two-stage full binary structure. A promising accuracy of 99.65%, with equally well recognition rates of over 99% throughout all type of ECG beats, has been achieved using the optimal feature set. Only 11 features are needed to attain this performance. The results demonstrate the effectiveness and efficiency of the proposed method for the computer-aided diagnosis of heart diseases based on ECG signals

ECG beat classification using particle swarm optimization and radial basis function neural network

The heart is one of the important parts of any human being. The heart produces electrical signals thus electrical signals are normally called as Electrocardiogram (ECG) signal. The Electrocardiogram signal is used for identifying the heart problems. The objective of this work is to implement an ANFIS algorithm for (ECG) signals classification. In this work, the classification is done using the ANFIS associated with back propagation algorithm. The ANFIS model is combination of adaptive capabilities with neural network the qualitative approach of fuzzy logic. The feature selection process is done before classification. Four types of ECG beats are collected from the PhysioBank databases. These heart signals are classified by four ANFIS classifiers. The fifth ANFIS classifier is used to get an improved diagnostic accuracy in the ECGs.

The occlusion in one of the coronary arteries of the heart leads to the cardiac ailment, myocardial infarction (MI). The localization of MI based on the investigation of the morphology of the multi-lead electrocardiogram (ECG) is the initial task for the diagnosis of this ailment. In this paper, the multiscale convolutional neural network is proposed for the automated localization of MI ailment from multi-lead electrocardiogram (ECG) beats. The Fourier-Bessel (FB) series expansion based empirical wavelet transform (EWT) with fixed order ranges is introduced for the multiscale analysis of multi-lead ECG beat. The FB spectrum of each lead ECG beat is segregated into contiguous segments using the fixed order ranges. Furthermore, the order ranges from these contiguous segments are used to design an empirical wavelet filter bank for the extraction of subband signals from each lead ECG beat. The convolutional neural network (CNN) is used for the classification of various categories of MI as anterior MI (AMI), anterio-lateral MI (ALMI), anterio-septal MI (ASMI), inferior MI (IMI), inferio-lateral MI (ILMI), inferio-posterio-lateral MI (IPLMI) and normal sinus rhythm (NSR). The experimental results reveal that the lower-order range subband signal coupled with CNN attains higher average accuracy values of 99.92%, 99.34%, 99.95%, 99.95%, 99.91%, and 99.86% respectively, for AMI, ALMI, ASMI, IMI, ILMI, and IPLMI classes. The subband signal of multi-lead ECG beats with order range of [1-26] is highly affected during various categories of MI heart disease, and this band signal has higher performance as compared to the existing MI localization approaches.

Automatic detection of life threatening abnormal beats in electrocardiogram (ECG) signal is of importance in many healthcare applications. The ECG beat signal variations in both shape and time impose great challenges to automatic detection tasks. To address those challenges and for high accuracy automatic detection, we present here a two stage abnormal beats detection algorithm. Normal and abnormal beat types are represented by templates which are selected from training data using clustering. Multiple templates for each beat type well represent the distribution of the data and allow nonlinearity of the discrimination. Critical features are extracted for both templates and incoming beat signals in discrete wavelet transform domain. The template matching is carried out using time invariant dynamic time warping (DTW). For those the DTW distance cannot provide sharp discrimination, an additional stage of verification is invoked to further check among those types having small distance with the incoming beat, based on ECG wave's time interval features. The algorithm is tested on a large data set (23 records containing 28114 normal beats and 4633 abnormal beats) and achieves an accuracy of 97.24%.

This paper introduces an efficient digital system design using hardware concepts to filter the Electrocardiogram (ECG) signal and to detect QRS complex (beats). The system implementation has been done using a Field Programmable Gate Array (FPGA) in two phases. In the first phase, Finite Impule Response (FIR) filters are designed for preprocessing and denoising the ECG signal. The filtered signal is then used as the input of the second phase to detect and classify the ECG beats. The entire system has been implemented on ALTERA DE II FPGA by desinging synthesizable finite state machines. The design has been tested on ECG waves from the MIT-BIH Arrhythmia database by windowing the signal and applying adaptive signal and noise theresholds in each window of processing. The hardware system has achieved an overall accuracy of 98% in the beat detection phase, while providing the detected beats and the classification of irregular heat-beat rates in real-time. The synthesized hardware of the ECG denoising and beat detection system yields reasonable hardware resources, making the system attractive to be eventually fabricated as a stand alone hardware system or integrated/embedded within a portable electronic device for monitoring patients’ conditions on a daily basis conveinently.

This work proposes an efficient method of Electrocardiogram (ECG) beats classification. ECG is an important biological signal indicating muscular activities of heart. Abnormalities in ECG beats reflect heart malfunctions. Four types of ECG beats including normal (N), left bundle branch block (LBBB), right bundle branch block (RBBB), and premature ventricular contractions (PVC) are intended to be classified. We take two-level wavelet decomposition to decompose ECG beats and extract appropriate features from sub bands of the decomposed signals. According to previous studies, wavelet coefficients seem to be sufficient features for solving the classification problems. However, they do not have suitable performance in high White Gaussian noise (WGN) added to ECG signals. Therefore, we compute fractal dimension using Katz' algorithm besides wavelet coefficients to complement our feature sets. Feature sets are classified by employing probabilistic neural network (PNN) which shows satisfactory results. Final results are indication of highly stable accuracy and high efficiency in different values of added WGN.

Robust mixture of experts modeling using the [formula omitted] distribution

Automatic ECG arrhythmia classification using dual tree complex wavelet based features

In this paper, the relationship between mixture of experts (ME) and the ensemble neural networks is discussed. We show that a special case of the ME reduces to the ensemble nets, and thus the weighting parameters of the ensemble nets can be obtained through the EM algorithm.

The Electrocardiogram (ECG) is one of the most commonly known biological signals. Traditionally ECG recordings are analysed in the time-domain by skilled physicians. However, pathological conditions may not always be obvious in the original time-domain signal. Fourier analysis provides frequency information but has the disadvantage that time characteristics will be lost. Wavelet analysis, which provides both time and frequency information, can overcome this limitation. Here a new method, the combination of wavelet analysis and feature vectors, is applied with the intent to investigate its suitability as a diagnostic tool. ECG signals with normal and abnormal beats were examined. There were two stages in analysing ECG signals: feature extraction and feature classification. To extract features from ECG signals, wavelet decomposition was first applied and feature vectors of normalised energy and entropy were constructed. These feature vectors were used to classify signals. The results showed that normal beats and abnormal beats composed different clusters in most cases. In conclusion, the combination of wavelet transform and feature vectors has shown potential in detecting abnormalities in an ECG recording. It was also found that normalised energy and entropy are features, which are suitable for classification of ECG signals.

Information geometry of the EM and em algorithms for neural networks

Parametrical modelling of a premature ventricular contraction ECG beat: Comparison with the normal case