Search

• A classification method of multiple cardiovascular diseases based on modified ECG. • Modified ECG, after EMD analysis, has better denoising performance than the raw ECG. • 1-D CNN learns the inherent features of the modified ECG better than the raw ECG. • Experimented on three public databases and the method outperforms existing methods. • The modified ECG signal is trained faster than any IMF and combinations of IMFs. Multiple cardiovascular disease classification from Electrocardiogram (ECG) signal is necessary for efficient and fast remedial treatment of the patient. This paper presents a method to classify multiple heart diseases using one dimensional deep convolutional neural network (CNN) where a modified ECG signal is given as an input signal to the network. Each ECG signal is first decomposed through Empirical Mode Decomposition (EMD) and higher order Intrinsic Mode Functions (IMFs) are combined to form a modified ECG signal. It is believed that the use of EMD would provide a broader range of information and can provide denoising performance. This processed signal is fed into the CNN architecture that classifies the record according to cardiovascular diseases using softmax regressor at the end of the network. It is observed that the CNN architecture learns the inherent features of the modified ECG signal better in comparison with the raw ECG signal. The method is applied on three publicly available ECG databases and it is found to be superior to other approaches in terms of classification accuracy. In MIT-BIH, St.-Petersberg, PTB databases the proposed method achieves maximum accuracy of 97.70%, 99.71%, and 98.24%, respectively. [ABSTRACT FROM AUTHOR]

Electrocardiogram (ECG) is a widely employed tool for the analysis of cardiac disorders. A clean ECG is often desired for proper treatment of cardiac ailments. However, in the real scenario, ECG signals are corrupted with various noises during acquisition and transmission. In this article, an efficient ECG denoising methodology using combined empirical mode decomposition (EMD) and adaptive switching mean filter (ASMF) is proposed. The advantages of both EMD and ASMF techniques are exploited to reduce the noises in the ECG signals with minimum distortion. Unlike conventional EMD based techniques, which reject the initial intrinsic mode functions (IMFs) or utilize a window based approach for reducing high-frequency noises, here, a wavelet based soft thresholding scheme is adopted for reduction of high-frequency noises and preservation of QRS complexes. Subsequently, an ASMF operation is performed to enhance the signal quality further. The ECG signals of standard MIT-BIH database are used for the simulation study. Three types of noises in particular white Gaussian noise, Electromyogram (EMG) and power line interference contaminate the test ECG signals. Three standard performance metrics namely output SNR improvement, mean square error, and percentage root mean square difference measure the efficacy of the proposed technique at various signal to noise ratio (SNR). The proposed denoising methodology is compared with other existing ECG denoising approaches. A detail qualitative and quantitative study and analysis indicate that the proposed technique can be used as an effective tool for denoising of ECG signals and hence can serve for better diagnostic in computer-based automated medical system. [ABSTRACT FROM AUTHOR]

Denoising of electrooculography (EOG) signals is a challenging task as the noise and signal share the same frequency band. This paper proposes a two-stage framework for denoising EOG signals. The first stage approach is based on preserving the nature of eye movements while the second stage is based on the nature of noise (Gaussian or not). In the first stage, denoising is carried out using one out of four filtering methods, each filter being optimal for a particular EOG pattern. The four methods used in the first stage are linear bandpass filtering, stationary wavelet transform (SWT), empirical mode decomposition (EMD) and median filtering. The Stage I framework selects the output that provides the highest estimated signal to noise ratio (SNR). In case, the Stage I filtering does not provide a significant SNR, the system uses Stage II filtering. In the second stage, we use two recursive state estimators, i.e. a Kalman filter and a particle filter for further denoising. The two-stage method is found to provide a better SNR as compared to a single stage method. [ABSTRACT FROM AUTHOR]

Deep learning has achieved promising results on a broad spectrum of tasks using an end-to-end approach, and domain-specific knowledge can be used to supplement it by either constraining the solution-space, or to transform data such that relevant signals are more influential during the training process. This is especially important in the context of continuous recordings or wearable-based monitoring, where signal quality may be constrained by hardware, such as a limited number of available electrodes, or by external perturbations. In this study, a global feature extraction method for multi-periodic signals is proposed, which effectively classifies multi-cycle normal and abnormal heartbeats. The abnormal heartbeats are classed as the Left Bundle Branch Block beat (LBBB), Right Bundle Branch Block beat (RBBB), and Paced beat (P). The signal analysis process of the electrocardiogram (ECG) recordings consist of three main stages: (I) segmentation of the ECG data and partitioning of the data set; (II) generation of an overall feature map representing the "heartbeat condition" based on Empirical Mode Decomposition (EMD); and (III) a classification stage for determining the patient's heartbeat condition. The extracted feature image is used to classify the heartbeat condition using a two-dimensional Convolutional Neural Network (CNN). This method is applied to the publicly available MIT-BIH arrhythmia database. Experimental results show that the reconstructed ECG features outperform use of the raw ECG signals. Under self-test conditions of 3 min of ECG signal, the Total Classification Accuracy (TCA) was approximately 99.01%, while the TCA without the proposed method was approximately 90.69%. • Necessary operations related to partial filtering and denoising can be omitted. • Preprocessing operations are reduced and the difficulty of preprocessing is reduced. • Overcome the shortcoming that instantaneous heart rate is easy to be misclassified. • Improve the classification accuracy of personal ECG. • Helps reduce the false positive rate of ECG recognition of wearable detection devices. [ABSTRACT FROM AUTHOR]

Deep learning has achieved promising results on a broad spectrum of tasks using an end-to-end approach, and domain-specific knowledge can be used to supplement it by either constraining the solution-space, or to transform data such that relevant signals are more influential during the training process. This is especially important in the context of continuous recordings or wearable-based monitoring, where signal quality may be constrained by hardware, such as a limited number of available electrodes, or by external perturbations. In this study, a global feature extraction method for multi-periodic signals is proposed, which effectively classifies multi-cycle normal and abnormal heartbeats. The abnormal heartbeats are classed as the Left Bundle Branch Block beat (LBBB), Right Bundle Branch Block beat (RBBB), and Paced beat (P). The signal analysis process of the electrocardiogram (ECG) recordings consist of three main stages: (I) segmentation of the ECG data and partitioning of the data set; (II) generation of an overall feature map representing the "heartbeat condition" based on Empirical Mode Decomposition (EMD); and (III) a classification stage for determining the patient's heartbeat condition. The extracted feature image is used to classify the heartbeat condition using a two-dimensional Convolutional Neural Network (CNN). This method is applied to the publicly available MIT-BIH arrhythmia database. Experimental results show that the reconstructed ECG features outperform use of the raw ECG signals. Under self-test conditions of 3 min of ECG signal, the Total Classification Accuracy (TCA) was approximately 99.01%, while the TCA without the proposed method was approximately 90.69%. • Necessary operations related to partial filtering and denoising can be omitted. • Preprocessing operations are reduced and the difficulty of preprocessing is reduced. • Overcome the shortcoming that instantaneous heart rate is easy to be misclassified. • Improve the classification accuracy of personal ECG. • Helps reduce the false positive rate of ECG recognition of wearable detection devices. [ABSTRACT FROM AUTHOR]

Microwave Doppler radar has received considerable attention as a non-contact form of measuring human respiration; in particular for long term monitoring. One of the main challenges in converting this into a viable application is to suppress or separate the artefacts and other interfering signals from the desired respiration signal using a less complex and practically feasible design for regular and potentially real time use. Existing systems either require complex experimental setups or multiple Doppler radar modules to achieve this. In this paper, we propose an approach based on EMD-ICA and approximate entropy ideas to systematically separate received Doppler shifted signal into distinct components and reconstruct the desired respiration pattern pertaining to respective physiological activity. Indeed this allows suppression of the undesirable artefacts and interference from other competing signals. Practical experiments confirmed comparable performance of the proposed method to the measurements obtained through chest straps which are widely used clinically for monitoring respiration. [ABSTRACT FROM AUTHOR]

• The proposed method presents a study with an EMD approach aimed at automatically diagnosing migraine from EEG signals. • Testing the classification performance for electrode-based migraine diagnosis with the proposed method. • Testing the distinctiveness of the obtained features by applying different statistical tests (Kruskal Wallis-Wilcoxon). • Comparison of the computational complexity of the proposed method with different decomposition methods. • Increasing classification performance in EEG-based migraine diagnosis by considering existing CAD studies. This study presents an Empirical Mode Decomposition (EMD) approach that aims to automatically detect migraine disease (MD) from electroencephalogram (EEG) recordings of migraine patient (MP) and healthy control (HC) subjects. First, the Multiscale Principal Component Analysis (MSPCA) method was applied to remove the noise on the raw EEG signals. Later, EEG signals were separated into intrinsic mode functions (IMF) components by EMD method. Statistical features were calculated and extracted from each IMF component. By applying the Kruskal Wallis (KW) test, the ability to distinguish these features in classification was tested. Classification performances are tested by classifying the features of each IMF component with a few leading ensemble algorithms. The highest classification accuracy of 92.47% was achieved by classifying the features of the IMF1 component with the Random Forest learning algorithm. At the last stage of the study, a comparative analysis with different time–frequency analysis methods is presented. As a result of the experimental comparison of our proposed method, it has been observed that it has a higher classification performance than other studies that detect EEG-based MD. With these aspects, our study reveals that it has the potential to be used as a computer aided diagnosis system that will support expert opinion in the detection of MD. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a new ECG denoising approach based on noise reduction algorithms in empirical mode decomposition (EMD) and discrete wavelet transform (DWT) domains. Unlike the conventional EMD based ECG denoising approaches that neglect a number of initial intrinsic mode functions (IMFs) containing the QRS complex as well as noise, we propose to perform windowing in the EMD domain in order to reduce the noise from the initial IMFs instead of discarding them completely thus preserving the QRS complex and yielding a relatively cleaner ECG signal. The signal thus obtained is transformed in the DWT domain, where an adaptive soft thresholding based noise reduction algorithm is employed considering the advantageous properties of the DWT compared to that of the EMD in preserving the energy in the presence of noise and in reconstructing the original ECG signal with a better time resolution. Extensive simulations are carried out using the MIT-BIH arrythmia database and the performance of the proposed method is evaluated in terms of several standard metrics. The simulation results show that the proposed method is able to reduce noise from the noisy ECG signals more accurately and consistently in comparison to some of the stateof-the-art methods. [Copyright &y& Elsevier]

The electrocardiogram (ECG) signal is popular and extensively used as a diagnostic tool for revealing several diseases related to the heart. Unfortunately, its sensitive to noises from various sources. Denoising these noises remains a hard and major challenge task, although there are some existing methods for denoising, they remain limited to a few types of noises with significant distortion after the filtering process. In this paper, we investigated a hybrid system that can deal with different types of noises, with different levels of SNR, conserving all the signal features with the minimum distortion. The hybrid system consists of a decomposition method followed by local means (LM) filtering. The best candidates used for the signal decomposition are the Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), Discrete Wavelet Transform (DWT), and Stationary Wavelet Transform (SWT). To demonstrate the effectiveness of the proposed method, several real ECG recordings and synthetic ECG infected by a different type of noise (i.e., baseline wander, muscle artifacts, electrode motion, mix noise, white noise, colored noise) with different levels of signal to noise ratios (24, 18, 6, 0, −6) dB are used in our numerical analysis. The performance evaluation of our proposed system is made using the following metrics: SNR improvement, signal-to-noise and distortion ratio (SINAD), the approximate entropy, and the fuzzy entropy. The experimental results emphasized that the Ensemble Empirical Mode Decomposition (EEMD) followed by local means (LM) outperforms all the comparison methods, in terms of all the metrics used, with few exceptions where the Empirical Mode Decomposition (EMD) followed by local means (LM) and Discrete Wavelet Transform (DWT) followed by local means (LM) performed well. [ABSTRACT FROM AUTHOR]

• The paper focuses on the need for muscle artifact detection for efficient analysis. • An Entropy based M-DDTW technique has been proposed for muscle artifact (MA) detection. • An accuracy of 95 % and a Performance Index of 90 % has been achieved in identifying MAs. • Reduces the computational time amd improves performance of EMD and EEMD based artifact extraction techniques. The exclusion of artifacts plays an indispensable role in the processing of Electroencephalographic (EEG) recordings. This work highlights one such inescapable artifactual event known as muscle artifacts (MA) and its detection methodology. These are high-frequency signals that are recurrently present in EEG and generally recorded via Electromyogram. The paper presents an entropy based Manhattan derivative dynamic time warping (M-DDTW) technique for MA epoch detection. Manhattan (City block) and Canberra distance have been proposed as the distance to be optimized by using them with dynamic time warping (DTW) and derivative dynamic time warping (DDTW) technique. The study reduces the computational time and improves the performance by utilizing entropy for reference generation and identifies the optimal threshold value for each technique. The results for the optimal threshold have been validated on the real EEG dataset. It was observed that the proposed entropy based M-DDTW technique exhibits the highest performance of 90 % and an accuracy of 95 % at an optimal threshold surpassing state of the art techniques. The testing of qualitative performance and time consumption has been done using traditional mode decomposition methods. The proposed Entropy based M-DDTW technique along with EEMD showed a noteworthy performance compared to other techniques. Overall the combination of entropy with time warping based local distance variation appears to be an adequate solution for muscle artifact detection. [ABSTRACT FROM AUTHOR]

Contactless measurement of cardio-vascular pulse acts an essential role in clinical medical sectors. Estimating cardio-vascular pulses based on continuous-wave (CW) Doppler radar in limited time while maintaining the accuracy is a challenging task. In this paper, we propose a signal processing method that combines empirical mode decomposition (EMD) and continuous wavelet transform (CWT) for a short time estimation of heart rate (HR) and inter-beat-interval from radar signals. We evaluate performance of the proposed method using 85 patients with dengue fever and 40 healthy subjects. Subsequently, the estimated contactless HR is compared to that of a commercial contact-type medical device. The result shows that the HR can be estimated within a period of 5 s with an accuracy of 96.2 ± 2.5%. The patients with dengue fever show an elevated HR and a decreased standard deviation of heartbeat interval (SDHI). Finally, linear discriminant analysis (LDA) is utilized with two parameters HR and SDHI to classify the dengue patients, achieving the sensitivity and specificity values of 86.3% and 86.9%, respectively. [ABSTRACT FROM AUTHOR]

• A new feature, instantaneous energy deviation cepstral coefficient (IEDCC) is proposed. • The average accuracy of 82% to 90% with vowels and 80% to 91% for words is achieved. • An increment of 20% in classification accuracy is achieved over the standard MFCC and acoustic features. • In dysarthria level prediction, highest correlation value of 0.75 with the vowel /o/ and 0.77 with the word /reina/ is obtained. Parkinson's disease (PD) is a progressive neurological disorder that mainly affects people in old age. Abnormality in the speech signals has been reported as a biomarker to detect PD. This study explores the use of Hilbert spectrum (HS) based features to model voice impairments in people affected by PD. The instantaneous energy deviation cepstral coefficient (IEDCC) is proposed. Statistical analyses show that the proposed feature is an effective and relevant biomarker for PD detection and evaluation of the dysarthria level in speech affected by PD. The capability of the proposed features to differentiate between PD and healthy people is evaluated upon five sustained vowels and ten isolated words from the standard PC-GITA database. The average accuracy of the proposed approach ranges from 82 % to 90 % with vowels, whereas for words the average accuracy ranges between 80 % and 91 %. Besides PD detection, the dysarthria level is evaluated according to the m-FDA scale. Spearman's correlation coefficients (ρ) are computed between the estimated m-FDA values and the original scores. Correlations of up to 0.75 are obtained with vowel/o/, while 0.77 is the highest correlation obtained with the word/reina/. The developed models are further validated with a separate and independent dataset. The classification accuracy in these additional recordings ranges between 50 % and 80 % with vowels and from 50 % to 82 % with words. The promising results obtained on the additional test set indicate that the proposed method is suitable to perform the automatic detection of PD speakers in real-world conditions. [ABSTRACT FROM AUTHOR]

• A new feature, instantaneous energy deviation cepstral coefficient (IEDCC) is proposed. • The average accuracy of 82% to 90% with vowels and 80% to 91% for words is achieved. • An increment of 20% in classification accuracy is achieved over the standard MFCC and acoustic features. • In dysarthria level prediction, highest correlation value of 0.75 with the vowel /o/ and 0.77 with the word /reina/ is obtained. Parkinson's disease (PD) is a progressive neurological disorder that mainly affects people in old age. Abnormality in the speech signals has been reported as a biomarker to detect PD. This study explores the use of Hilbert spectrum (HS) based features to model voice impairments in people affected by PD. The instantaneous energy deviation cepstral coefficient (IEDCC) is proposed. Statistical analyses show that the proposed feature is an effective and relevant biomarker for PD detection and evaluation of the dysarthria level in speech affected by PD. The capability of the proposed features to differentiate between PD and healthy people is evaluated upon five sustained vowels and ten isolated words from the standard PC-GITA database. The average accuracy of the proposed approach ranges from 82 % to 90 % with vowels, whereas for words the average accuracy ranges between 80 % and 91 %. Besides PD detection, the dysarthria level is evaluated according to the m-FDA scale. Spearman's correlation coefficients (ρ) are computed between the estimated m-FDA values and the original scores. Correlations of up to 0.75 are obtained with vowel/o/, while 0.77 is the highest correlation obtained with the word/reina/. The developed models are further validated with a separate and independent dataset. The classification accuracy in these additional recordings ranges between 50 % and 80 % with vowels and from 50 % to 82 % with words. The promising results obtained on the additional test set indicate that the proposed method is suitable to perform the automatic detection of PD speakers in real-world conditions. [ABSTRACT FROM AUTHOR]