| Summary: | This thesis presents a novel method for the discrimination between various levels of anesthetic-induced
unconsciousness (i.e. hypnosis) in the context of intra-operative anesthesia. This is achieved by means of
the wavelet analysis of the patient's Electroencephalogram (EEG).
While monitors of hypnosis are already available, they rely on complex signal processing algorithms, and
in some instances, on multivariate statistical analysis of a large quantity of data. To date, the BIS® monitor
(Aspect Medical Systems, Inc., MA) provides the most accepted solution. It displays an index calculated
based on bispectral analysis, and has been the subject of many clinical trials over the past 5 years. As such,
it is widely known and accepted by the medical community, particularly in the United States.
Wavelet analysis can be viewed as a generalization of Fourier analysis that introduces time localization in
addition to frequency decomposition of a signal. This major advantage of wavelets makes them particularly
suitable for the analysis of non-stationary signals such as the EEG. In this thesis, we present a simple
wavelet-based technique that calculates, in real time, an index of hypnosis (so-called WAV Index) based on
a patient's electroencephalogram. Wavelet analysis significantly reduces the computational complexity to
perform this task. Further, neither a large number of patients nor an extensive amount of clinical EEG data
is needed for the index derivation.
In addition, a technique for the de-noising of the electroencephalogram to remove ocular artifacts is
proposed. These artifacts, which result from ocular activity (eye blinks and movements), strongly perturb
the EEG signal by superimposing large spikes and step-like patterns. Hypnotic states such as the awake state
and light sedation are typically corrupted by these waveforms. The solution proposed successfully removes
these artifacts from the EEG, thus providing a clean signal for the wavelet analysis. This technique further
works in the case of head movement artifacts.
Finally, clinical validation has been carried out to assess the performance of the proposed index. While
being particularly well-correlated to the bispectral index, the proposed index has a faster response to fast
transients and it is more consistent in terms of patient intra-variability.
|
|---|
