Estimation of Event-Related Signals from the Brain

Measurement of the electrical responses from the brain, the ElectroEncephaloGram (EEG) is a major technique for analyzing and understanding the processes of the brain. The two main reasons for this are the out-standing time resolution and the cheap cost compared to other techniques, e.g., Magnetic Resonance Imaging and other related methods which often require costly equipments. Robust spectral analysis techniques are extremely valuable in the analysis of EEG signals. The Evoked Potential (EP) is the electrical response from the brain caused by some external stimuli, e.g., visual or auditive, and is usually heavily disturbed by the spontaneous brain activity. The EPs repeat with essentially the same form and the clinical method used today is averaging over several stimuli. This procedure reduces the disturbance from the EEG as this activity can be modeled as noise. We have applied different methods for extracting the parameters for estimation of the P300-wave, a special type of EP, e.g., the Prony method. In another project, a new algorithm was developed to reduce the number of averages for estimation of another EP, the Nb-wave, where the amplitude and latency of the Nb-wave are utilized for measurement of depth of anesthesia. This algorithm is also patented in several countries.

More recently we have focused on induced potentials, which are the phase-varying responses of certain events. These responses differ from the EPs which in contrary could be described as the in-phase responses of events. Usual averaging techniques over different realizations are not appropriate for induced responses as the phase-variation will give a bad average. Instead, averaging over the time-varying power indicates the performance. The frequency content is usually estimated by successively averaged subspectra from different time epochs. Events of short duration will be difficult to detect and the onset and offset time of those events will be misinterpreted. Optimal time-frequency kernels and corresponding multitapers for a class of locally stationary processes (MW-LSP) where the covariance function is determined by two one-dimensional functions have been studied and evaluated. Examples are shown for different parameter values tuning the stationarity of the LSP (larger c-value corresponds to more non-stationary process). The aim is to estimate a local transient in the EEG, evoked by a flickering light of 9 Hz and lasting 1 s with start at 0. The results show the response of the flickering light and also an evoked stronger alpha-activity between 10 and 11 Hz and around 2 s.

Using robust time-frequency techniques, we have estimated the responses during the second year of life of children watching films with actors performing different communicative behavior. This work is also connected to the Linnaeus Grant "Thinking in Time: Cognition, Communication and learning" at Lund University, where it is studied how humans, in particular children, learn concepts and to understand the words of a language. Neural responses, to video clips displaying different types of non-linguistic communication between two actors, were recorded. Results from investigations of language learning indicate that there may be important developments in this domain in the second half of the second year and as example of the differences that is found, the grand average spectrogram estimates corresponding to imperative and declarative gestures compared to neutral gestures. The grand average spectrogram of the usual evoked activity, estimated by time averages, (SOM), are located around 1-2 Hz. However, studying the grand average of the single spectrograms (MOS) show the strongest difference activity at around 6Hz for declarative gestures. To support this research, an evaluation on the optimal window lengths and number of windows of multiple window time-frequency analysis methods is recently published.