Isolation of Fetal ECG Signals from Abdominal ECG Using Wavelet Analysis

ObjectiveMonitoring the heartbeat of the fetus during pregnancy is a vital part in determining their health. Current fetal heart monitoring techniques lack the accuracy in fetal heart rate monitoring and features acquisition, resulting in diagnostic medical issues. The demand for a reliable method of non-invasive fetal heart monitoring is of high importance.MethodElectrocardiogram (ECG) is a method of monitoring the electrical activity produced by the heart. The extraction of the fetal ECG (FECG) from the abdominal ECG (AECG) is challenging since both ECGs of the mother and the baby share similar frequency components, adding to the fact that the signals are corrupted by white noise. This paper presents a method of FECG extraction by eliminating all other signals using AECG. The algorithm is based on attenuating the maternal ECG (MECG) by filtering and wavelet analysis to find the locations of the FECG, and thus isolating them based on their locations. Two signals of AECG collected at different locations on the abdomens are used. The ECG data used contains MECG of a power of five to ten times that of the FECG.ResultsThe FECG signals were successfully isolated from the AECG using the proposed method through which the QRS complex of the heartbeat was conserved, and heart rate was calculated. The fetal heart rate was 135 bpm and the instantaneous heart rate was 131.58 bpm. The heart rate of the mother was at 90 bpm with an instantaneous heart rate of 81.9 bpm.ConclusionThe proposed method is promising for FECG extraction since it relies on filtering and wavelet analysis of two abdominal signals for the algorithm. The method implemented is easily adjusted based on the power levels of signals, giving it great ease of adaptation to changing signals in different biosignals applications.     

Arrhythmic Heartbeat Classification Using 2D Convolutional Neural Networks

BackgroundElectrocardiogram (ECG) is a method of recording the electrical activity of the heart and it provides a diagnostic means for heart-related diseases. Arrhythmia is any irregularity of the heartbeat that causes an abnormality in the heart rhythm. Early detection of arrhythmia has great importance to prevent many diseases. Manual analysis of ECG recordings is not practical for quickly identifying arrhythmias that may cause sudden deaths. Hence, many studies have been presented to develop computer-aided-diagnosis (CAD) systems to automatically identify arrhythmias.MethodsThis paper proposes a novel deep learning approach to identify arrhythmias in ECG signals. The proposed approach identifies arrhythmia classes using Convolutional Neural Network (CNN) trained by two-dimensional (2D) ECG beat images. Firstly, ECG signals, which consist of 5 different arrhythmias, are segmented into heartbeats which are transformed into 2D grayscale images. Afterward, the images are used as input for training a new CNN architecture to classify heartbeats.ResultsThe experimental results show that the classification performance of the proposed approach reaches an overall accuracy of 99.7%, sensitivity of 99.7%, and specificity of 99.22% in the classification of five different ECG arrhythmias. Further, the proposed CNN architecture is compared to other popular CNN architectures such as LeNet and ResNet-50 to evaluate the performance of the study.ConclusionsTest results demonstrate that the deep network trained by ECG images provides outstanding classification performance of arrhythmic ECG signals and outperforms similar network architectures. Moreover, the proposed method has lower computational costs compared to existing methods and is more suitable for mobile device-based diagnosis systems as it does not involve any complex preprocessing process. Hence, the proposed approach provides a simple and robust automatic cardiac arrhythmia detection scheme for the classification of ECG arrhythmias.     

A Comparison of ECG Signal Pre-processing Using FrFT,FrWT and IPCA for Improved Analysis

ObjectiveElectrocardiogram (ECG) is a diagnostic tool for recording electrical activities of the human heart non-invasively. It is detected by electrodes placed on the surface of the skin in a conductive medium. In medical applications, ECG is used by cardiologists to observe heart anomalies (cardiovascular diseases) such as abnormal heart rhythms, heart attacks, effects of drug dosage on subject's heart and knowledge of previous heart attacks. Recorded ECG signal is generally corrupted by various types of noise/distortion such as cardiac (isoelectric interval, prolonged depolarization and atrial flutter) or extra cardiac (respiration, changes in electrode position, muscle contraction and power line noise). These factors hide the useful information and alter the signal characteristic due to low Signal-to-Noise Ratio (SNR). In such situations, any failure to judge the ECG signal correctly may result in a delay in the treatment and harm a subject (patient) health. Therefore, appropriate pre-processing technique is necessary to improve SNR to facilitate better treatment to the subject. Effects of different pre-processing techniques on ECG signal analysis (based on R-peaks detection) are compared using various Figures of Merit (FoM) such as sensitivity (Se), accuracy (Acc) and detection error rate (DER) along with SNR.MethodsIn this research article, a new fractional wavelet transform (FrWT) has been proposed as a pre-processing technique in order to overcome the disadvantages of other existing commonly used techniques viz. wavelet transform (WT) and the fractional Fourier transform (FrFT). The proposed FrWT technique possesses the properties of multiresolution analysis and represents signal in the fractional domain which consists of representation in terms of rotation of signals in the time–frequency plane. In the literature, ECG signal analysis has been improvised using statistical pre-processing techniques such as principal component analysis (PCA), and independent component analysis (ICA). However, both PCA and ICA are prone to suffer from slight alterations in either signal or noise, unless the basis functions are prepared with a worldwide set of ECG. Independent Principal Component Analysis (IPCA) has been used to overcome this shortcoming of PCA and ICA. Therefore, in this paper three techniques viz. FrFT, FrWT and IPCA are selected for comparison in pre-processing of ECG signals.ResultsThe selected methods have been evaluated on the basis of SNR, Se, Acc and DER of the detected ECG beats. FrWT yields the best results among all the methods considered in this paper; 34.37dB output SNR, 99.98% Se, 99.96% Acc, and 0.036% DER. These results indicate the quality of biology-related information retained from the pre-processed ECG signals for identifying different heart abnormalities.ConclusionCorrect analysis of the acquired ECG signal is the main challenge for cardiologist due to involvement of various types of noises (high and low frequency). Twenty two real time ECG records have been evaluated based on various FoM such as SNR, Se, Acc and DER for the proposed FrWT and existing FrFT and IPCA preprocessing techniques. Acquired real-time ECG database in normal and disease situations is used for the purpose. The values of FoMs indicate high SNR and better detection of R-peaks in a ECG signal which is important for the diagnosis of cardiovascular disease. The proposed FrWT outperforms all other techniques and holds both analytical attributes of the actual ECG signal and alterations in the amplitudes of various ECG waveforms adequately. It also provides signal portrayals in the time-fractional-frequency plane with low computational complexity enabling their use practically for versatile applications.     

The interpretation of abdominal wall muscle recruitment strategies change when the electrocardiogram (ECG) is removed from the electromyogram (EMG)

The purpose of this study was to determine the effect of the ECG artifact on low-level trunk muscle activation amplitudes and assess the effectiveness of two methods used to remove the ECG. Simulations were performed and percent error in root mean square (RMS) amplitudes were calculated from uncontaminated and contaminated EMG signals at various ECG to EMG ratios. Two methods were used to remove the ECG: (1) filtering by adaptive sampling (FAS) and (2) Butterworth high pass filter at 30 Hz (BW-30 Hz HPF). The percent error was also calculated between the ECG removed and the uncontaminated EMG RMS amplitudes. Next, the BW-30 Hz HPF method was used to remove the ECG from 3-bilateral external oblique (EO) muscle sites collected from 30 healthy subjects performing a one handed lift and replace task. Two separate ANOVA models assessed the effects of ECG on the statistical interpretation of EO recruitment strategies. One model included EMG data that contained the ECG and the other model included EMG data after the ECG was removed. Large percent errors were observed when the ECG was not removed. These errors increased with larger ECG to EMG ratios. Both removal methods reduced the errors to below 10%, but the BW-30 Hz HPF method was more time efficient in removing the ECG artifact. Different statistical findings were observed among the muscle sites for the ECG contaminated model compared to the ECG removed model, which resulted in different conclusions concerning neuromuscular control.     

Electrocardiogram Data Compression Techniques for Cardiac Healthcare Systems: A Methodological Review

Objective: Globally, cardiovascular diseases (CVDs) are one of the most leading causes of death. In medical screening and diagnostic procedures of CVDs, electrocardiogram (ECG) signals are widely used. Early detection of CVDs requires acquisition of longer ECG signals. It has triggered the development of personal healthcare systems which can be used by cardio-patients to manage the disease. These healthcare systems continuously record, store, and transmit the ECG data via wired/wireless communication channels. There are many issues with these systems such as data storage limitation, bandwidth limitation and limited battery life. Involvement of ECG data compression techniques can resolve all these issues.Method: In the past, numerous ECG data compression techniques have been proposed. This paper presents a methodological review of different ECG data compression techniques based on their experimental performance on ECG records of the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) arrhythmia database.Results: It is observed that experimental performance of different compression techniques depends on several parameters. The existing compression techniques are validated using different distortion measures.Conclusion: This study elaborates advantages and disadvantages of different ECG data compression techniques. It also includes different validation methods of ECG compression techniques. Although compression techniques have been developed very widely but the validation of compression methods is still a prospective research area to accomplish an efficient and reliable performance.     

Elimination of electrocardiogram contamination from electromyogram signals: An evaluation of currently used removal techniques.

Trunk electromyographic signals (EMG) are often contaminated with heart muscle electrical activity (ECG) due to the proximity of the collection sites to the heart and the volume conduction characteristics of the ECG through the torso. Few studies have quantified ECG removal techniques relative to an uncontaminated EMG signal (gold standard or criterion measure), or made direct comparisons between different methods for a given set of data. Understanding the impacts of both untreated contaminated EMG and ECG elimination techniques on the amplitude and frequency parameters is vital given the widespread use of EMG. The purpose of this study was to evaluate four groups of current and commonly used techniques for the removal of ECG contamination from EMG signals. ECG recordings at two intensity levels (rest and 50% maximum predicted heart rate) were superimposed on 11 uncontaminated biceps brachii EMG signals (rest, 7 isometric and 3 isoinertial levels). The 23 removal methods used were high pass digital filtering (finite impulse response (FIR) using a Hamming window, and fourth-order Butterworth (BW) filter) at five cutoff frequencies (20, 30, 40, 50, and 60 Hz), template techniques (template subtraction and an amplitude gating template), combinations of the subtraction template and high pass digital filtering, and a frequency subtraction/signal reconstruction method. For muscle activation levels between 10% and 25% of maximum voluntary contraction, the template subtraction and BW filter with a 30 Hz cutoff were the two best methods for maximal ECG removal with minimal EMG distortion. The BW filter with a 30 Hz cutoff provided the optimal balance between ease of implementation, time investment, and performance across all contractions and heart rate levels for the EMG levels evaluated in this study.     

Empirical Mode Decomposition and Wavelet Transform Based ECG Data Compression Scheme

ObjectiveIn health-care systems, compression is an essential tool to solve the storage and transmission problems. In this regard, this paper reports a new electrocardiogram (ECG) data compression scheme which employs sifting function based empirical mode decomposition (EMD) and discrete wavelet transform.MethodEMD based on sifting function is utilized to get the first intrinsic mode function (IMF). After EMD, the first IMF and four significant sifting functions are combined together. This combination is free from many irrelevant components of the signal. Discrete wavelet transform (DWT) with mother wavelet ‘bior4.4’ is applied to this combination. The transform coefficients obtained after DWT are passed through dead-zone quantization. It discards small transform coefficients lying around zero. Further, integer conversion of coefficients and run-length encoding are utilized to achieve a compressed form of ECG data.ResultsCompression performance of the proposed scheme is evaluated using 48 ECG records of the MIT-BIH arrhythmia database. In the comparison of compression results, it is observed that the proposed method exhibits better performance than many recent ECG compressors. A mean opinion score test is also conducted to evaluate the true quality of the reconstructed ECG signals.ConclusionThe proposed scheme offers better compression performance with preserving the key features of the signal very well.     

Paroxysmal atrial fibrillation recognition based on multi-scale wavelet <Emphasis Type="Italic">α</Emphasis>-entropy

Background

This study proposed an effective method based on the wavelet multi-scale α-entropy features of heart rate variability (HRV) for the recognition of paroxysmal atrial fibrillation (PAF). This new algorithm combines wavelet decomposition and non-linear analysis methods. The PAF signal, the signal distant from PAF, and the normal sinus signals can be identified and distinguished by extracting the characteristic parameters from HRV signals and analyzing their quantification indexes. The original ECG signals for QRS detection and HRV signal extraction are first processed. The features from the HRV signals are extracted as feature vectors using the wavelet multi-scale entropy. A support vector machine-based classifier is used for PAF prediction.

Results

The performance of the proposed method in predicting PAF episodes is evaluated with 100 signals from the MIT-BIT PAF prediction database. With regard to the dynamics and uncertainty of PAF signals, our proposed method obtains the values of 92.18, 94.88, and 89.48% for the evaluation criteria of correct rate, sensitivity, and specificity, respectively.

Conclusions

Our proposed method presents better results than the existing studies based on time domain, frequency domain, and non-linear methods. Thus, our method shows considerable potential for clinical monitoring and treatment.

     

An Efficient Automated Algorithm for Distinguishing Normal and Abnormal ECG Signal

ObjectiveThe present study aims to simulate an alarm system for online detecting normal electrocardiogram (ECG) signals from abnormal ECG so that an individual's heart condition can be accurately and quickly monitored at any moment, and any possible serious dangers can be prevented.Materials and methodsFirst, the data from Physionet database were used to analyze the ECG signal. The data were collected equally from both males and females, and the data length varied between several seconds to several minutes. The heart rate variability (HRV) signal, which reflects heart fluctuations in different time intervals, was used due to the low spatial accuracy of ECG signal and its time constraint, as well as the similarity of this signal with the normal signal in some diseases. In this study, the proposed algorithm provided a return map as well as extracted nonlinear features of the HRV signal, in addition to the application of the statistical characteristics of the signal. Then, artificial neural networks were used in the field of ECG signal processing such as multilayer perceptron (MLP) and support vector machine (SVM), as well as optimal features, to categorize normal signals from abnormal ones.ResultsIn this paper, the area under the curve (AUC) of the ROC was used to determine the performance level of introduced classifiers. The results of simulation in MATLAB medium showed that AUC for MLP and SVM neural networks was 89.3% and 94.7%, respectively. Also, the results of the proposed method indicated that the more nonlinear features extracted from the ECG signal could classify normal signals from the patient.ConclusionThe ECG signal representing the electrical activity of the heart at different time intervals involves some important information. The signal is considered as one of the common tools used by physicians to diagnose various cardiovascular diseases, but unfortunately the proper diagnosis of disease in many cases is accompanied by an error due to limited time accuracy and hiding some important information related to this signal from the physicians' vision leading to the risks of irreparable harm for patients. Based on the results, designing the proposed alarm system can help physicians with higher speed and accuracy in the field of diagnosing normal people from patients and can be used as a complementary system in hospitals.     

An Improved SSA-Based Technique for EMG Removal from ECG

Background

The electrocardiogram (ECG) signals provide important information about the heart electrical activities in medical and diagnostic applications. This signal may be contaminated by different types of noises. One of the noise types which has a considerable overlap with the ECG signals in frequency domain is electromyogram (EMG). Among the exciting approaches for de-noising the ECG signals, those based on singular spectrum analysis (SSA) are popular.

A novel approach for removing ECG interferences from surface EMG signals using a combined ANFIS and wavelet

In recent years, the removal of electrocardiogram (ECG) interferences from electromyogram (EMG) signals has been given large consideration. Where the quality of EMG signal is of interest, it is important to remove ECG interferences from EMG signals. In this paper, an efficient method based on a combination of adaptive neuro-fuzzy inference system (ANFIS) and wavelet transform is proposed to effectively eliminate ECG interferences from surface EMG signals. The proposed approach is compared with other common methods such as high-pass filter, artificial neural network, adaptive noise canceller, wavelet transform, subtraction method and ANFIS. It is found that the performance of the proposed ANFIS–wavelet method is superior to the other methods with the signal to noise ratio and relative error of 14.97 dB and 0.02 respectively and a significantly higher correlation coefficient (p < 0.05).     

1D-CADCapsNet: One dimensional deep capsule networks for coronary artery disease detection using ECG signals

PurposeCardiovascular disease (CVD) is a leading cause of death globally. Electrocardiogram (ECG), which records the electrical activity of the heart, has been used for the diagnosis of CVD. The automated and robust detection of CVD from ECG signals plays a significant role for early and accurate clinical diagnosis. The purpose of this study is to provide automated detection of coronary artery disease (CAD) from ECG signals using capsule networks (CapsNet).MethodsDeep learning-based approaches have become increasingly popular in computer aided diagnosis systems. Capsule networks are one of the new promising approaches in the field of deep learning. In this study, we used 1D version of CapsNet for the automated detection of coronary artery disease (CAD) on two second (95,300) and five second-long (38,120) ECG segments. These segments are obtained from 40 normal and 7 CAD subjects. In the experimental studies, 5-fold cross validation technique is employed to evaluate performance of the model.ResultsThe proposed model, which is named as 1D-CADCapsNet, yielded a promising 5-fold diagnosis accuracy of 99.44% and 98.62% for two- and five-second ECG signal groups, respectively. We have obtained the highest performance results using 2 s ECG segment than the state-of-art studies reported in the literature.Conclusions1D-CADCapsNet model automatically learns the pertinent representations from raw ECG data without using any hand-crafted technique and can be used as a fast and accurate diagnostic tool to help cardiologists.     

基于多尺度快速样本熵与随机森林的心电图分析

目的:探讨基于多尺度快速样本熵与随机森林的心电图分析方法对常见心律失常(房性早搏、室性早搏)的自动诊断的可行性和有效性。方法:利用不同心律失常疾病的心电信号存在复杂性差异的特点,通过多尺度熵计算心电信号在不同尺度下的样本熵值以组成特征向量;利用kd树提高多尺度熵的计算效率,增强算法的实时性。利用训练样本的特征向量构建随机森林分类器,再根据众多决策树的分类结果结合投票原则确定测试样本心律失常疾病的类型。结果:本文提出的心电图分析方法能够有效地识别正常心律、房性早搏(APB)及室性早搏(VPB),平均识别准确率达到91.60%。结论:本文提出的心电图分析方法对常见心律失常(APB,VPB)具有较高的识别准确率及临床实用价值。     

High resolution ECG analysis by an improved signal averaging method and comparison with a beat-to-beat approach

The aim of this paper is to describe the analysis of a high resolution ECG recorded from the body surface. Standard signal averaging techniques are improved by using a new time delay estimation method which leads to a better alignment accuracy of P and T waves. A second method uses adaptive identification to achieve a beat by beat fine ECG estimation. Information provided by the two methods allows a better interpretation of low and very low level signals.     

Nonlinear analysis of continuous ECG during sleep I. Reconstruction

In recent years evidence has accumulated that ECG signals are of a nonlinear nature. It has been recognized that strictly periodic cardiac rhythms are not accompanied by healthy conditions but, on the contrary, by pathological states. Therefore, the application of methods from nonlinear system theory for the analysis of ECG signals has gained increasing interest. Crucial for the application of nonlinear methods is the reconstruction (embedding) of the time series in a phase space with appropriate dimension. In this study continuous ECG signals of 12 healthy subjects recorded during different sleep stages were analysed. Proper embedding dimension was determined by application of two techniques – the false nearest neighbours method and the saturation of the correlation dimension. Results for the ECG signals were compared with findings for simulated data (quasiperiodic dynamics, Lorenz data, white noise) and for phase randomized surrogates. Findings obtained with the two approaches suggest that embedding dimensions from 6 to 8 may be regarded as suitable for the topologically proper reconstruction of ECG signals. Received: 7 June 1999 / Accepted in revised form: 10 December 1999     

Effect of square pulsed magnetic field exposure on growth kinetics of Dickeya solani

Abstract

The effect of square pulsed magnetic signals on the growth of Dickeya solani bacterium was studied. Three different frequency windows in a range up to 50?Hz were selected for exposure to determine the frequencies causing maximum bacteriostatic and bactericidal effects. Furthermore, the cellular morphological changes under most inhibitory conditions were studied and bacterial pathogenicity was examined. The obtained data showed the most inhibitory frequency that caused an inhibition by 65% and delayed the cellular growth by more than 8?h is 5?Hz. The morphological studies of exposed bacterium cells showed cellular abnormalities and fragmentations and loss of membrane outer surface charges. The pathogenicity test exposed a significant decrease in the infection reached 68%. In conclusion, the present study developed better disease strategy for controlling the bacterium of D. solani in an efficient and safe manner. Moreover, the applicability of using pulsed electromagnetic signals may have a good role in bacterial inhibition.     

Online Classification of Multiple Motor Imagery Tasks Using Filter Bank Based Maximum-a-Posteriori Common Spatial Pattern Filters

ObjectiveThe main objective of this paper is to propose a novel technique, called filter bank maximum a-posteriori common spatial pattern (FB-MAP-CSP) algorithm, for online classification of multiple motor imagery activities using electroencephalography (EEG) signals. The proposed technique addresses the overfitting issue of CSP in addition to utilizing the spectral information of EEG signals inside the framework of filter banks while extending it to more than two conditions.Materials and methodsThe classification of motor imagery signals is based upon the detection of event-related de-synchronization (ERD) phenomena in the μ and β rhythms of EEG signals. Accordingly, two modifications in the existing MAP-CSP technique are presented: (i) The (pre-processed) EEG signals are spectrally filtered by a bank of filters lying in the μ and β brainwave frequency range, (ii) the framework of MAP-CSP is extended to deal with multiple (more than two) motor imagery tasks classification and the spatial filters thus obtained are calculated for each sub-band, separately. Subsequently, the most imperative features over all sub-bands are selected and un-regularized linear discriminant analysis is employed for classification of multiple motor imagery tasks.ResultsPublicly available dataset (BCI Competition IV Dataset I) is used to validate the proposed method i.e. FB-MAP-CSP. The results show that the proposed method yields superior classification results, in addition to be computationally more efficient in the case of online implementation, as compared to the conventional CSP based techniques and its variants for multiclass motor imagery classification.ConclusionThe proposed FB-MAP-CSP algorithm is found to be a potential / superior method for classifying multi-condition motor imagery EEG signals in comparison to FBCSP based techniques.     

Investigating the Effect of Flickering Frequency Pair and Mother Wavelet Selection in Steady-State Visually-Evoked Potentials on Two-Command Brain-Computer Interfaces

IntroductionSteady-state visually evoked potentials (SSVEPs) have become popular in brain-computer interface (BCI) applications in addition to many other applications on clinical neuroscience (neurodegenerative disorders, schizophrenia, epilepsy, etc.), cognitive (visual attention, working memory, brain rhythms, etc.), and use of engineering researches. Among available methods to measure brain activities, SSVEPs have advantages like higher information transfer rate, simplicity in structure, and short training time. SSVEP-based BCIs use flickering stimuli at different frequencies to discriminate distinct commands in real life. Some features are extracted from the SSVEP signals before these commands are classified. The wavelet transform (WT) has attracted researchers among feature extraction methods since it utilizes the non-stationary signals well. In the WT, a sample function (named mother wavelet) represents the SSVEP signal in both time and frequency domains. Unfortunately, there is no universal mother wavelet function that fits all the signals. Therefore, choosing an appropriate mother wavelet function may be a challenge in WT-related studies. Although there are such studies in three- and seven-command SSVEP-based studies, there is no study for two-command systems in our knowledge.Materials and MethodsIn this study, two user commands flickered at the combinations of seven different frequencies were tested to determine which frequency pairs give the highest performance. For this purpose, three well-known wavelet features (energy, entropy, and variance) were calculated for each of derived EEG frequency bands from the discrete WT coefficients of SSVEP signals. The WT was repeated for six different mother wavelet functions (Haar, Db4, Sym4, Coif1, Bior3.5, and Rbior2.8). Then, four feature sets (every three features, and all together) were applied to seven commonly-used machine learning algorithms (Decision Tree, Discriminant Analysis, Logistic Regression, Naive Bayes, Support Vector Machines, Nearest Neighbors, and Ensemble Classifiers).Results and DiscussionWe achieved 100% accuracies among these 3,528 runs (7 classifiers x 4 feature sets x 6 mother wavelets x 21 flickering frequency pairs) using the mother wavelet function of Haar and the Ensemble Learner classifier. The highest classifier performances are 100% when two commands have the flickering frequency pairs of (6.0 and 10 Hz), (6.5 and 8.2 Hz), or (6.5 and 10.0 Hz).ConclusionWe obtained three main outcomes from this study. First, the most representative mother wavelet function was Haar, while the worst one was Symlet 4. Second, the Ensemble Learner classifier gave the maximum classifier performance in a two-command SSVEP-based BCI system. Besides, two user commands from SSVEP should be one of the frequency pairs of (6.0 and 10.0 Hz), (6.5 and 8.2 Hz), and (6.5 and 10.0 Hz) to achieve the maximum accuracy.     

Classification of Motor Imagery EEG Based on Time-Domain and Frequency-Domain Dual-Stream Convolutional Neural Network

Background and objectiveAn important task of the brain-computer interface (BCI) of motor imagery is to extract effective time-domain features, frequency-domain features or time-frequency domain features from the raw electroencephalogram (EEG) signals for classification of motor imagery. However, choosing an appropriate method to combine time domain and frequency domain features to improve the performance of motor imagery recognition is still a research hotspot.MethodsIn order to fully extract and utilize the time-domain and frequency-domain features of EEG in classification tasks, this paper proposed a novel dual-stream convolutional neural network (DCNN), which can use time domain signal and frequency domain signal as the inputs, and the extracted time-domain features and frequency-domain features are fused by linear weighting for classification training. Furthermore, the weight can be learned by the DCNN automatically.ResultsThe experiments based on BCI competition II dataset III and BCI competition IV dataset 2a showed that the model proposed by this study has better performance than other conventional methods. The model used time-frequency signal as the inputs had better performance than the model only used time-domain signals or frequency-domain signals. The accuracy of classification was improved for each subject compared with the models only used one signals as the inputs.ConclusionsFurther analysis shown that the fusion weight of different subject is specifically, adjusting the weight coefficient automatically is helpful to improve the classification accuracy.