Ghasem Azemi

Principles of time-frequency feature extraction for change detection in non-stationary signals

This paper considers the general problem of detecting change in non-stationary signals using features observed in the time-frequency (t,f) domain, obtained using a class of quadratic time-frequency distributions (QTFDs). The focus of this study is to propose a methodology to define new (t,f) features by extending time-only and frequency-only features to the joint (t,f) domain for detecting changes in non-stationary signals. The (t,f) features are used as a representative subset characterizing the status of the observed non-stationary signal. Change in the signal is then reflected as a change in the (t,f) features. This (t,f) approach is applied to the problem of detecting abnormal brain activity in newborns (e.g. seizure) using measurements of the EEG for diagnosis and prognosis. In addition, a pre-processing stage for detecting artifacts in EEG signals for signal enhancement is studied and implemented separately. Overall results indicate that, in general, the (t,f) approach results in an improved performance in detecting artifacts and seizures in newborn EEG signals as compared to time-only or frequency-only features. HighlightsWe propose (t,f) based features for detecting change in nonstationary signals.We use the features to detect seizures and artifacts in newborn EEGs.The features result in an improved performance in detecting seizures and artifacts.Performance of (t,f) features depends on the type of time-frequency distribution.

A methodology for time-frequency image processing applied to the classification of non-stationary multichannel signals using instantaneous frequency descriptors with application to newborn EEG signals

This article presents a general methodology for processing non-stationary signals for the purpose of classification and localization. The methodology combines methods adapted from three complementary areas: time-frequency signal analysis, multichannel signal analysis and image processing. The latter three combine in a new methodology referred to as multichannel time-frequency image processing which is applied to the problem of classifying electroencephalogram (EEG) abnormalities in both adults and newborns. A combination of signal related features and image related features are used by merging key instantaneous frequency descriptors which characterize the signal non-stationarities. The results obtained show that, firstly, the features based on time-frequency image processing techniques such as image segmentation, improve the performance of EEG abnormalities detection in the classification systems based on multi-SVM and neural network classifiers. Secondly, these discriminating features are able to better detect the correlation between newborn EEG signals in a multichannel-based newborn EEG seizure detection for the purpose of localizing EEG abnormalities on the scalp.

Measuring Time-Varying Information Flow in Scalp EEG Signals: Orthogonalized Partial Directed Coherence

This study aimed to develop a time-frequency method for measuring directional interactions over time and frequency from scalp-recorded electroencephalographic (EEG) signals in a way that is less affected by volume conduction and amplitude scaling. We modified the time-varying generalized partial directed coherence (tv-gPDC) method, by orthogonalization of the strictly causal multivariate autoregressive model coefficients, to minimize the effect of mutual sources. The novel measure, generalized orthogonalized PDC (gOPDC), was tested first using two simulated models with feature dimensions relevant to EEG activities. We then used the method for assessing event-related directional information flow from flash-evoked responses in neonatal EEG. For testing statistical significance of the findings, we followed a thresholding procedure driven by baseline periods in the same EEG activity. The results suggest that the gOPDC method 1) is able to remove common components akin to volume conduction effect in the scalp EEG, 2) handles the potential challenge with different amplitude scaling within multichannel signals, and 3) can detect directed information flow within a subsecond time scale in nonstationary multichannel EEG datasets. This method holds promise for estimating directed interactions between scalp EEG channels that are commonly affected by the confounding impact of mutual cortical sources.

A review of time-frequency matched filter design with application to seizure detection in multichannel newborn EEG

This paper presents a novel design of a time-frequency (t-f) matched filter as a solution to the problem of detecting a non-stationary signal in the presence of additive noise, for application to the detection of newborn seizure using multichannel EEG signals. The solution reduces to two possible t-f approaches that use a general formulation of t-f matched filters (TFMFs) based on the Wigner-Ville and cross Wigner-Ville distributions, and a third new approach based on the signal ambiguity domain representation; referred to as Radon-ambiguity detector. This contribution defines a general design formulation and then implements it for newborn seizure detection using multichannel EEG signals. Finally, the performance of different TFMFs is evaluated for different t-f kernels in terms of classification accuracy using real newborn EEG signals. Experimental results show that the detection method which uses TFMFs based on the cross Wigner-Ville distribution outperforms other approaches including the existing TFMF-based ones. The results also show that TFMFs which use high-resolution kernels such as the modified B-distribution, achieve higher detection accuracies compared to the ones which use other reduced-interference t-f kernels.

Time-frequency signal and image processing of non-stationary signals with application to the classification of newborn EEG abnormalities

This paper presents an introduction to time-frequency (T-F) methods in signal processing, and a novel approach for EEG abnormalities detection and classification based on a combination of signal related features and image related features. These features which characterize the non-stationary nature and the multi-component characteristic of EEG signals, are extracted from the T-F representation of the signals. The signal related features are derived from the T-F representation of EEG signals and include the instantaneous frequency, singular value decomposition, and energy based features. The image related features are extracted from the T-F representation considered as an image, using T-F image processing techniques. These combined signal and image features allow to extract more information from a signal. The results obtained on newborn and adult EEG data, show that the image related features improve the performance of the EEG seizure detection in classification systems based on multi-SVM classifier.

Robust estimation of highly-varying nonlinear instantaneous frequency of monocomponent signals using a lower-order complex-time distribution

This paper proposes an approach for robust estimation of highly-varying nonlinear instantaneous frequency (IF) in monocomponent nonstationary signals. The proposed method is based on a lower order complex-time distribution (CTD), derived by using the idea of complex-time differentiation of the instantaneous phase. Unlike other existing TFDs in the same framework, the proposed TFD is an order-free distribution which alleviates the subtractive cancellation error in IF estimation. The approach is applied to highly nonstationary monocomponent signals. Performance of the numerical implementation is compared with three existing IF estimation methods using three simulated signals. Noise analysis is also performed to evaluate the robustness of the method in presenfdece of additive noise at signal to noise ratio (SNR) varying from -10dB to 20dB. Results show that the proposed method outperforms the other methods at lower SNR and works reasonably well for the noiseless case.

A time-frequency based approach for generalized phase synchrony assessment in nonstationary multivariate signals

This paper proposes a new approach to estimate the phase synchrony among nonstationary multivariate signals using the linear relationships between their instantaneous frequency (IF) laws. For cases where nonstationary signals are multi-component, a decomposition method like multi-channel empirical mode decomposition (MEMD) is used to simultaneously decompose the multi-channel signals into their intrinsic mode functions (IMFs). We then apply the Johansen method on the IF laws to assess the phase synchrony within multivariate nonstationary signals. The proposed approach is validated first using multi-channel synthetic signals. The method is then used for quantifying the inter-hemispheric EEG asynchrony during ictal and inter-ictal periods using a newborn EEG seizure/non-seizure database of five subjects. For this application, pair-wise phase synchrony measures may not be able to account for phase interactions between multiple channels. Furthermore, the classical definition of phase synchrony, which is based on the rational relationships between phases, may not reveal the hidden phase interdependencies caused by irrational long-run relationships. We evaluate the performance of the proposed method using the differentiation of unwrapped phase as well as other IF estimation techniques. The results obtained on newborn EEG signals confirm that the generalized phase synchrony within EEG channels increases significantly during ictal periods. A statistically consistent phase coupling is also observed within the non-seizure segments supporting the concept of constant inter-hemispheric connectivity in the newborn brain during inter-ictal periods.

Improved characterization of HRV signals based on instantaneous frequency features estimated from quadratic time–frequency distributions with data-adapted kernels

The analysis of heart rate variability (HRV) provides a non-invasive tool for assessing the autonomicregulation of cardiovascular system. Quadratic time–frequency distributions (TFDs) have been used toaccount for the non-stationarity of HRV signals, but their performance is affected by cross-terms. Thisstudy presents an improved type of quadratic TFD with a lag-independent kernel (LIK-TFD) by introducinga new parameter defined as the minimal frequency distance among signal components. The resultingTFD with this LIK can effectively suppress the cross-terms while maintaining the time–frequency (TF)resolution needed for accurate characterization of HRV signals. Results of quantitative and qualitativetests on both simulated and real HRV signals show that the proposed LIK-TFDs outperform other TFDscommonly used in HRV analysis. The findings of the study indicate that these LIK-TFDs provide morereliable TF characterization of HRV signals for extracting new instantaneous frequency (IF) based clinicallyrelated features. These IF based measurements shown to be important in detecting perinatal hypoxicinsult – a severe cause of morbidity and mortality in newborns.

Improving the classification of newborn EEG time-frequency representations using a combined time-frequency signal and image approach

This paper presents new time-frequency (T-F) features to improve the classification of non-stationary signals such as EEG signals. Previous methods were based only on signal features that were derived from the instantaneous frequency and energies of EEG signals in different spectral sub-bands. This paper includes new features that are based on T-F image descriptors which are extracted from the T-F representation considered as an image, using T-F image processing techniques. The results obtained on newborn EEG data, show that the use of image related-features with signal based-features improve the performance of the newborn EEG seizure detection and classification when using multi-SVM classifiers. These results allow the possibility of improving health outcomes for sick babies by early intervention on the basis of the results of the classification of newborn EEG abnormalities.

EEG-based automatic epilepsy diagnosis using the instantaneous frequency with sub-band energies

This paper presents a novel approach for classifying the electroencephalogram (EEG) signals as normal or abnormal. This method uses features derived from the instantaneous frequency (IF) and energies of EEG signals in different spectral sub-bands. Results of applying the method to a database of real signals reveal that, for the given classification task, the selected features consistently exhibit a high degree of discrimination between the EEG signals collected from healthy and epileptic patients. The analysis of the effect of window length used during feature extraction indicates that features extracted from EEG segments as short as 5 seconds achieve a high average total accuracy of 95.3%.