Waveform decomposition: Identification and modeling of evoked potential components
Abstract
The evoked potential (EP) is modeled as the sum of gaussian pulses whose amplitudes, latencies and width may vary in a random fashion. The null hypothesis stating that the latencies and amplitudes of the EP components are gaussian distributed random variables is tested with the Kolmogorov-Smirnov goodness-of-fit test. A decomposition algorithm for resolving the EP into its components is introduced. The algorithm generates auxiliary functions from experimental data. In the limit, each one of these auxiliary functions will contain a delta function and a second function consisting of a linear superposition of the gaussian pulses. At each step of the process, a new auxiliary function is created, and at each step, different parameters of the complex waveform are extracted. The different components of each of the auxiliary functions have a shorter duration than those in the original waveform. By successively reducing the time duration, the degree of interaction and amount of overlap between independent adjacent components is decreased and the resolution between the different peaks of the waveform is improved. The improvement being measured by the ability of the algorithm to resolve what appears as a peak into two components. The main innovation of the model is that it allows for dependent peaks to be formed from the interaction of adjacent components of the same polarity. The Segment-Eliminate-Search algorithm for identifying dependent components in single evoked potentials is developed. The algorithm segments the signal and studies triplets of peaks. Properties of the derivative of gaussian pulses are used to eliminate the first peak in the triplet. The function generated is then searched for the existence of a peak in a predefined region. A new ensemble of evoked potentials is generated. The single responses of this new ensemble are generated by applying the Segment-Eliminate-Search algorithm to each single evoked potential estimate obtained after using the Multichannel Time Varying Filter. The segments in each single potential which contain dependent peaks are removed. Cubic spine interpolation is used to estimate new values for the segments removed.
Degree
Ph.D.
Advisors
Aunon, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.