A Kalman filter based methodology for EEG spike enhancement

Unit of Medical Technology and Intelligent Information Systems, Department of Computer Science, University of Ioannina, GR 45110 Ioannina, Greece.
Computer Methods and Programs in Biomedicine (Impact Factor: 1.9). 03/2007; 85(2):101-8. DOI: 10.1016/j.cmpb.2006.10.003
Source: PubMed

ABSTRACT In this work, we present a methodology for spike enhancement in electroencephalographic (EEG) recordings. Our approach takes advantage of the non-stationarity nature of the EEG signal using a time-varying autoregressive model. The time-varying coefficients of autoregressive model are estimated using the Kalman filter. The results show considerable improvement in signal-to-noise ratio and significant reduction of the number of false positives.

Download full-text


Available from: Alexandros T Tzallas, Aug 12, 2015
1 Follower
  • Source
    • "Many approaches have been proposed to identify time-varying autoregressive TVAR models. While traditional adaptive parameter estimation algorithms, for example the recursive least squares (RLS) and least mean squares (LMS) can be applied to track time-varying trends [15]–[16], these algorithms can often produce lagged tracking of time varying parameters. Fast transversal recursive instrumental variable (FTRIV) and generalized least mean squares (GLMS) [17], are proposed for the estimation of AR non Gaussian processes. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this paper, we present an adaptive spectrum estimation method for non-stationary Biomedical Signals. The algorithm is based on time-varying autoregressive (TVAR) modeling where the time varying parameters are estimated by Kalman filtering. The algorithm generates adaptively an estimate of the power spectral density (PSD) at each time instant. A comparison was made with the recursive least squares (RLS) method, the main feature of the proposed approach is the capability of the Kalman filter that enables tracking smooth and sharp changes in the time varying process parameters. Furthermore, it provides better time-frequency resolution and gives a good spectral peak matching. Simulation studies and applications on real EEG data show that the proposed algorithm can provide important transient information on the inherent dynamics of non-stationary biomedical processes.
    The Open Electrical & Electronic Engineering Journal 08/2015; 2(4):59-67.
  • Source
    • "They may also be categorized by the features they used: morphological features [28] or time–frequency ones [22]. Most of the spike detection methods have an enhancement stage that generates an output signal in which the distinction between the spikes and the noise is increased by some filtering methods such as Wavelet Transform [29] [27], matched filters [30] or Kalman filter [31]. At this stage, the output signal is used in a decision procedure in order to extract the spike peak times. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Denoising is an important preprocessing stage in some ElectroEncephaloGraphy (EEG) applications. For this purpose, Blind Source Separation (BSS) methods, such as Independent Component Analysis (ICA) and Decorrelated and Colored Component Analysis (DCCA), are commonly used. Although ICA and DCCA-based methods are powerful tools to extract sources of interest, the procedure of eliminating the effect of sources of non-interest is usually manual. It should be noted that some methods for automatic selection of artifact sources after BSS methods exist, although they imply a training supervised step. On the other hand, in cases where there are some a priori information about the subspace of interest, semi-blind source separation methods can be used to denoise EEG signals. Among them the Generalized EigenValue Decomposition (GEVD) and Denoising Source Separation (DSS) are two well-known semi-blind frameworks that can be used with a priori information on the subspace of interest. In this paper, we compare the ICA and DCCA-based methods, namely CoM2 and SOBI, respectively, with GEVD and DSS in the application of extracting the epileptic activity from noisy interictal EEG data. To extract a priori information required by GEVD and DSS, we propose a series of preprocessing stages including spike peak detection, extraction of exact time support of spikes and clustering of spikes involved in each source of interest. The comparison of these four methods in terms of performance and numerical complexity shows that CoM2 give better performance for very low SNR values but require visual inspection to select the sources of interest. For higher SNR values, GEVD and DSS based approaches give similar results but with lower numerical complexity and without requiring a visual selection of the sources of interest.
    IRBM 11/2014; 36(1). DOI:10.1016/j.irbm.2014.10.002 · 0.38 Impact Factor
  • Source
    • "These parameters are the state transition matrix A, the covariance of the state noise Q, the variance of the driving noise , and the initial conditions, . The parameters θ can be tuned based on some empirical knowledge (Oikonomou et al., 2007) or define a function of parameters and perform optimization to obtain the optimal values of the parameters like the EM algorithm (Khan & Dutt, 2007). In the next section the EM algorithm is described to tune the parameters θ. "
    [Show abstract] [Hide abstract]
    ABSTRACT: In this chapter the Kalman Smoother, with or without the EM algorithm, has been used for the processing of the EEG signal in two cases, epileptic form spike identification and ERD/ERD analysis. Use of the Kalman Smoother forces to some simplifications of the model. This is performed in order to decrease the number of parameters which must be tuned. Based on the assumptions that the state transition matrix is the identity and the covariance is diagonal with the same element on the diagonal, there is one parameter to be tuned, the variance of the state noise. The value of this parameter defines how smooth or rough will be the evolution of states, in our case the TVAR coefficients. Large values of variance indicate rough estimates for the TVAR coefficients. This has as a result a noisy time varying spectrum. Small values indicated smooth estimates for the TVAR coefficients and hence a smooth time varying spectrum. The value of this parameter depends on the problem. In the case where we expect that the time varying spectrum is smooth, a small value for the variance of the state noise is preferable. However, the parameters can be estimated based on some optimization procedure like the EM algorithm. The EM algorithm provides with estimates of the parameters. So the tuning of the parameters is done automatically based on the dataset, without manual settings. This fact permits the use of full covariance for the state noise and a general transition state matrix. As a consequence the model is more flexible because of the different types of state noise. We observe that the Kalman Smoother with EM provides with smoother estimates than using the Kalman Smoother alone. This happens because the first approach can capture the patterns of the signal more accurately. In the estimation of the IF in the spike problem it is observed that the IF starts to increase before the appearance of the spike. Also, in the ERD/ERS analysis we observe that the IF is modulated when some events take place on the experiment, like the sound at t=2sec which denotes the beginning of the trial. In both problems we observe that the IF is a good measure to track changes in EEG activity.
    Kalman Filter Recent Advances and Applications, 04/2009; , ISBN: 978-953-307-000-1
Show more