Motion artifact is observed in electroencephalogram (EEG) signals during the acquisition. The elimination of this type of artifact using various signal processing approaches is considered a pre-processing task for different neural information processing applications. In this paper, the wavelet domain optimized Savitzky-Golay (WOSG) filtering approach was proposed for the removal of motion artifacts from EEG signals. The multiscale analysis of the EEG signals using discrete wavelet transform (DWT) produces sub-band signals at different scales. Motion artifact is a low-frequency artifact that appears in the approximation sub-band signal. The optimized SG filter was applied to the motion artifact intermixed approximation sub-band signal, and the cleaned approximation sub-band signal was evaluated based on the subtraction of the optimized SG filter output from the motion artifact intermixed sub-band signal. The filtered EEG signal was computed based on the addition of cleaned approximation sub-band signal with other sub-band signals of contaminated EEG. The proposed WOSG filtering approach was evaluated using EEG recordings from various publicly available databases. Measures such as the mean absolute error in power spectral density (MAE-PSD) of δ-band between contaminated and cleaned EEG signal, ΔSNR, percentage change in correlation coefficients (η), and mutual information (MI) were used to quantify the performance of the proposed filtering approach. The results revealed that the proposed WOSG filtering approach had superior denoising performance with the average ΔSNR, η, and MAE-PSD values of 30:59 dB, 68:76dB/Hz in comparison to the multi-resolution total variation (MTV) (ΔSNR as 29:12 dB, η as 68:560:0365 dB/Hz) and other existing methods. The approach had the average MI values of 4:152, and 4:103, and average MAE-PSD values of 0:276 dB/Hz, and 0:256 dB/Hz for δ-bands of EEG signals recorded during standing and walking conditions.

Electroencephalogram (EEG) is a diagnostic test, and it measures the entire brain’s electrical activity. The EEG signals have been used in many applications such as the diagnosis of neurological abnormalities, the brain-computer interface (BCI), the detection of sleep-related pathologies, etc. The EEG signal is contaminated with ocular artifact during the acquisition, and the filtering of this artifact is indeed required for efficient processing of this signal. In this work, we have proposed a method for the removal of ocular artifacts from the EEG signal. The Fourier-Bessel series expansion based empirical wavelet transform (FBSE-EWT) is used for the extraction of EEG rhythms namely, δ rhythm, θ rhythm, α rhythm, β rhythm and γ rhythm sub-signals from the ocular artifact contaminated EEG signal. The enhanced local polynomial (LP) approximation based total variation (TV) filtering is applied over the contaminated δ rhythm to obtain both LP and TV components. The filtered δ rhythm sub-signal is obtained based on the subtraction of both LP and TV components from the contaminated δ rhythm sub-signal. The filtered EEG signal is evaluated by combining the filtered δ rhythm with θ rhythm, α rhythm, β rhythm, and γ rhythm sub-signals. The energy ratio of the δ rhythm and the mean absolute error (MAE) in the power spectral density (PSD) values for all other rhythms are used as the performance metrics for the evaluation of the proposed method. The experimental results reveal that the proposed method has a better performance with a minimum average MAE in PSD value of 0.029 for α rhythm as compared to other existing techniques.

The categorization of sleep stages helps to diagnose different sleep-related ailments. In this paper, an entropy-based information–theoretic approach is introduced for the automated categorization of sleep stages using multi-channel electroencephalogram (EEG) signals. This approach comprises of three stages. First, the decomposition of multi-channel EEG signals into sub-band signals or modes is performed using a novel multivariate projection-based fixed boundary empirical wavelet transform (MPFBEWT) filter bank. Second, entropy features such as bubble and dispersion entropies are computed from the modes of multi-channel EEG signals. Third, a hybrid learning classifier based on class-specific residuals using sparse representation and distances from nearest neighbors is used to categorize sleep stages automatically using entropy-based features computed from MPFBEWT domain modes of multi-channel EEG signals. The proposed approach is evaluated using the multi-channel EEG signals obtained from the cyclic alternating pattern (CAP) sleep database. Our results reveal that the proposed sleep staging approach has obtained accuracies of 91.77%, 88.14%, 80.13%, and 73.88% for the automated categorization of wake vs. sleep, wake vs. rapid eye movement (REM) vs. Non-REM, wake vs. light sleep vs. deep sleep vs. REM sleep, and wake vs. S1-sleep vs. S2-sleep vs. S3-sleep vs. REM sleep schemes, respectively. The developed method has obtained the highest overall accuracy compared to the state-of-art approaches and is ready to be tested with more subjects before clinical application.

The neurological pathology which occurs due to the disturbance of the nerve cell activity and causing recurrent seizures is called epilepsy. In medical practice, the localization of the epileptogenic area or region in the brain is the primary task for the effectiveness of the epilepsy surgery. The epileptogenic region is identified based on the presence of focal electroencephalogram (EEG) signals during recording. Therefore, the classification of focal (FL) and non-focal (NFL) classes of EEG channels is the prerequisite to identify the epileptogenic regions in the brain. In this paper, a hybrid approach based on the combination of the band or rhythm specific Fourier-Bessel series expansion domain empirical wavelet transform (FBSE-EWT) filter bank and sparse autoencoder (SAE) based support vector machine (SAE-SVM) network is proposed for the categorization of FL and NFL types of EEG channels. The rhythms such as δ, θ, α, β, and γ are obtained from the EEG signal of each channel using FBSE-EWT filter bank. The SAE-SVM network classifies the FL and NFL categories of EEG channels directly from the rhythms. The results demonstrate that the proposed hybrid approach has 100% accuracy for the classification of FL and NFL types using the δ-rhythms of EEG signals for both channels. The approach extracts learnable features in the SAE stage, and these features have higher performance as compared to the existing features for the categorization of FL and NFL types of EEG channels.

Myocardial infarction (MI) occurs due to the decrease in the blood flow into one part of the heart, and it further causes damage to the heart muscle. The 12-channel electrocardiogram (ECG) has been widely used to detect and localize MI pathology in clinical studies. The vectorcardiogram (VCG) is a 3-channel recording system used to measure the heart’s electrical activity in sagittal, transverse, and frontal planes. The VCG signals have advantages over the 12-channel ECG to localize posterior MI pathology. Detection and localization of MI using VCG signals are vital in clinical practice. This paper proposes a multi-channel multi-scale two-stage deep-learning-based approach to detect and localize MI using VCG signals. In the first stage, the multivariate variational mode decomposition (MVMD) decomposes the three-channel-based VCG signal beat into five components along each channel. The multi-channel multi-scale VCG tensor is formulated using the modes of each channel of VCG data, and it is used as the input to the deep convolutional neural network (CNN) to classify MI and normal sinus rhythm (NSR) classes. In the second stage, the multi-class deep CNN is used for the categorization of anterior MI (AMI), anterior-lateral MI (ALMI), anterior-septal MI (ASMI), inferior MI (IMI), inferior-lateral MI (ILMI), inferior-posterior-lateral (IPLMI) classes using MI detected multi-channel multi-scale VCG instances from the first stage. The proposed approach is developed using the VCG data obtained from a public database. The results reveal that the approach has obtained the accuracy, sensitivity, and specificity values of 99.58%, 99.18%, and 99.87%, respectively, for MI detection. Moreover, for MI localization, we have obtained the overall accuracy value of 99.86% in the second stage for our proposed network. The proposed approach has demonstrated superior classification performance compared to the existing VCG signal-based MI detection and localization techniques.