Lin-Sen Pon

The bi-directional spike detection in EEG using mathematical morphology and wavelet transform

Epileptic EEG often contains abnormal spiky activity which is diagnostically important. It is proposed that the multi-resolution wavelet transform with mathematical morphology can be used to detect and extract this activity. Differentiating the geometrical characteristics between spikes and normal EEG activity, the process extracts the target patterns from the EEG data in the multi-resolution domains. Morphological analysis utilizes analytic operations based on a pre-defined structuring element (SE) targeted to specific signal features. In our case the SE is defined as a disk to measure the difference in smoothness between the two components. Discrete wavelet transforms are applied to construct the processed signal. The multi-resolution property of the wavelet transform adapts well to the time-invariant nature of the signal. Combining mathematical morphology and wavelet transforms, this method successfully separates the background activity and transient phenomenon from epileptic EEG. Although the morphological operation is a non-linear process, we show that, with the selected structuring element, this approach has ability to detect both positive and negative going spikes identically.

Application of time frequency and wavelet analysis to the diagnosis of epilepsy

Wavelet transforms and time-frequency distributions are powerful techniques for analysis of nonstationary biomedical signals. This paper investigates three applications of these techniques to multichannel electroencephalography (EEG) for the diagnosis of epilepsy. Wavelet transforms are utilized to detect the onset of seizures at different sites of subdural electrodes, and to extract spike patterns from EEG data recorded from the scalp. Time-frequency distributions are applied to characterize the early activity of seizures.

Analysis of heart rate changes associated with ictal activity

The authors examined the EKG signal and its relationship to seizure activity. The EKG data was divided into three periods: before, during and after ictal activity based on visual and behavior classification of the data. Traditional indices, such as the mean and variance, of the R-R intervals of heart beats are evaluated to compare the variation in these three periods. The spectra of the R-R intervals are investigated to determine a ratio between the highest variation (HV) and the lowest variation (LV). This ratio shows how fast the R-R Intervals change during these three periods. Nonlinear time-frequency analysis is also used to evaluate heart rate changes before and after ictal activity. The time-frequency analysis shows that the heart rate increases during the an ictal event and decreases back towards a normal range after the ictus.

Spike Separation from EEG/MEG Data Using Morphological Filter and Wavelet Transform

In the analysis of epileptic electroencephalographic (EEG) and magnetoencephalography (MEG) data, spike separation is diagnostically important because localization of epileptic focus often depends on accurate extraction of spiky activity from the raw data. In this paper, we present a method to automatically extract spikes using the wavelet transform combined with morphological filtering based on a circular structuring element. Our experimental results have shown that this method is highly effective in spike separation. Comparisons with the wavelet, bandpass filter, empirical mode decomposition (EMD), and independent component analysis (ICA) methods show that the new method is more effective in estimating both spike amplitudes and locations

POCS supperresolution image reconstruction using wavelet transform

Research interest in multi-frame supperresolution has risen substantially in recent years and it is expected to have wide applications to medicine. This paper presents a modified projection onto convex sets (POCS) superresolution method based on the wavelet transform. Based on the principle of POCS, an iterative procedure is proposed to extract information hidden in a group of low resolution images to update the corresponding high frequency band of the reference image, thus augmenting the individual low-resolution image to a high resolution image. The simulation results show that the new proposed algorithm is effective in constructing high-resolution medical images.

A medical EEG/video multimedia content description system

Neurological diagnostic information management for clinical applications requires significant collaboration among clinicians and integration of data in different forms, such as textual records, monitoring video, recorded physiological waveforms, and diagnostic images. In medical settings, these multimedia data are often generated in large amounts and are routinely compared and examined. In order to manage this type of information efficiently, an advanced information management system is investigated which handles not only the traditional textual and pictorial information, but also manages audio and video. This information system utilizes the newly developed MPEG-7 standard (multimedia content description interface standard) which provides a context-based platform for our application. In addition, MPEG-7 allows medical information to be accessed remotely and securely through the Internet.

Signal recovery from the approximation component in the non-downsampled wavelet transform

It is well known that a signal can be perfectly reconstructed from its wavelet-decomposed components: an approximation component and a set of detail components. Can a signal be recovered from its approximation component without detail components? This paper gives an answer to this question using a non-downsampled wavelet transform. Our experiments and analyses show that a signal can be recovered from its approximation coefficients solely by performing the non-downsampled wavelet transform iteratively. The results from the 2-level and 4-level wavelet transforms show that the recovered signal converges to the original signal as the number of iteration increases.

Interictal spike analysis using stochastic point process

Interictal spikes are important indicators of epileptic focus (foci). The spiky events in EEG waveforms recorded from different regions of the brain provide important information about the dynamic transitions of epileptic activity. We extract sequences of spikes, called spike trains, from individual EEG channels and evaluate them using the stochastic cross-correlogram, where the time of occurrence of each spike in a single channel is statistically correlated to the times of occurrences of spikes in other channels. A probability density function (pdf) is constructed which provides a distribution of the time intervals between the appearances of spike train events across a pair of channels. The time interval corresponding to the maximum of the pdf indicates a likelihood of latency between the two spiky events during the specified observation period. The higher the peak value, the stronger the cross-correlation between the two spike trains. We have applied the cross-correlogram analysis to subdural EEC data recorded from multiple electrodes spread over different regions of the brain. We calculated the maximum peak interval of the cross-correlogram between every pair of spike trains to observe the dynamical transitions of the interictal activity in the temporal and spatial domain. Our results show that, by inspecting the dynamic developments of spike events in these domains, a powerful tool is obtained for the diagnosis of epileptic focus (foci)

Analyzing EEG scalp current density using the Hjorth estimator

Scalp current density (SCD) measurements can be used to identify sources and sinks of electrical current on the scalp. We derive a SCD estimator from the Laplacian equation and compare it to the Hjorth estimator. It is clarified that the Hjorth estimator is the weighted average of the directional differences while the Laplacian estimator represents the summation of the second derivatives.