Objective: This paper investigates the multivariate oscillatory nature of electroencephalogram (EEG) signals in adaptive frequency scales for epileptic seizure detection. Methods: The empirical wavelet transform (EWT) has been explored for the multivariate signals in order to determine the joint instantaneous amplitudes and frequencies in signal adaptive frequency scales. The proposed multivariate extension of EWT has been studied on multivariate multicomponent synthetic signal, as well as on multivariate EEG signals of Children's Hospital Boston-Massachusetts Institute of Technology (CHB-MIT) scalp EEG database. In a moving-window-based analysis, 2-s-duration multivariate EEG signal epochs containing five automatically selected channels have been decomposed and three features have been extracted from each 1-s part of the 2-s-duration joint instantaneous amplitudes of multivariate EEG signals. The extracted features from each oscillatory level have been processed using a proposed feature processing step and joint features have been computed in order to achieve better discrimination of seizure and seizure-free EEG signal epochs. Results: The proposed detection method has been evaluated over 177 h of EEG records using six classifiers. We have achieved average sensitivity, specificity, and accuracy values as 97.91%, 99.57%, and 99.41%, respectively, using tenfold cross-validation method, which are higher than the compared state of art methods studied on this database. Conclusion: Efficient detection of epileptic seizure is achieved when seizure events appear for long duration in hours long EEG recordings. Significance: The proposed method develops time-frequency plane for multivariate signals and builds patient-specific models for EEG seizure detection.

This paper analyzes the underlying complexity and non-linearity of electroencephalogram (EEG) signals by computing a novel multi-scale entropy measure for the classification of seizure, seizure-free and normal EEG signals. The quality factor (Q) based multi-scale entropy measure is proposed to compute the entropy of the EEG signal in different frequency-bands of interest. The Q -based entropy (QEn) is computed by decomposing the signal with the tunable-Q wavelet transform (TQWT) into the number of sub-bands and estimating K-nearest neighbor (K-NN) entropies from various sub-bands cumulatively. The optimal selection of Q and the redundancy parameter (R) of TQWT showed better robustness for entropy computation in the presence of high- and low-frequency components. The extracted features are fed to the support vector machine (SVM) classifier with the wrapper-based feature selection method. The proposed method has achieved accuracy of 100% in classifying normal (eyes-open and eyes-closed) and seizure EEG signals, 99.5% in classifying seizure-free EEG signals (from the hippocampal formation of the opposite hemisphere of the brain) from seizure EEG signals and 98% in classifying seizure-free EEG signals (from the epileptogenic zone) from seizure EEG signals, respectively, using the SVM classifier. We have also achieved classification accuracies of 99% and 98.6% in classifying seizure versus non-seizure EEG signals and the individual three classes, namely normal, seizure-free and seizure EEG signals, respectively. The performance measure of the proposed multi-scale entropy has been found to be comparable with the existing state of the art epileptic EEG signals classification methods studied using the same database.

Myocardial infarction (MI) is also called the heart attack, and it results in the death of heart muscle cells due to the lacking in the supply of oxygen and other nutrients. The early and accurate detection of MI using the 12-lead electrocardiogram (ECG) is helpful in the clinical standard for saving the lives of the patients suffering from this pathology. This paper proposes a novel approach for the detection of MI pathology using the multiresolution analysis of 12-lead ECG signals. The approach is based on the use of Fourier- Bessel series expansion based empirical wavelet transform (FBSE-EWT) for the time-scale decomposition of 12-lead ECG signals. For each lead ECG signal, nine subband signals are evaluated using FBSE-EWT. The statistical features such as the kurtosis, the skewness, and the entropy are evaluated from the subband signals of each ECG lead. The deep neural network (DNN) such as the deep layer least square support vector machine (DL-LSSVM) which is formulated using the hidden layers of sparse auto-encoders and the LSSVM is used for the detection of MI from the feature vector of 12-lead ECG. The experimental results demonstrate that the combination of FBSE-EWT based entropy features and DL-LSSVM has the mean accuracy, the mean sensitivity and the mean specificity values of 99.74%, 99.87%, and 99.60%, respectively for the detection of MI. The accuracy value of the proposed method is improved by more than 3% as compared to the wavelet-based features for the detection of MI.

This work proposes a novel multivariate-multiscale approach for computing the spectral and temporal entropies from the multichannel electroencephalogram (EEG) signal. This facilitates the recognition of three human emotions: positive, neutral, and negative. The proposed approach is based on the application of the Fourier-Bessel series expansion based empirical wavelet transform (FBSE-EWT). We have extended the existing FBSE-EWT method for multichannel signals and derived FBSE-EWT based multivariate Hilbert marginal spectrum (MHMS) for computing spectral Shannon and K-nearest neighbor (KNN) entropies. The multivariate FBSE-EWT decomposes the multichannel EEG signals into narrow band subband signals. The multiscaling operation adapted in the spectral domain is based on the selection of successive joint instantaneous amplitude and frequency functions of the subband signals. On the other hand, the time domain multiscale K-NN entropy is computed from the cumulatively added multidimensional subband signals. The extracted spectral and temporal entropy features are smoothed and fed to the sparse autoencoder based random forest (ARF) classifier architecture for emotion classification. The proposed approach is tested using multichannel EEG signals available in a public database (SJTU emotion EEG dataset (SEED)). The bivariate EEG signals from different channel pairs with distinct spatial locations over the scalp are considered as input to our proposed system. The obtained overall classification accuracy of 94.4% reveals that the proposed approach is useful in classifying human emotions. The method is also validated using DREAMER emotion EEG public database. The method outperforms the existing state-of-the-art methods evaluated in these databases.