Stadnyk M. Information technology of the steady-state visual evoked potentials analysis in the ophthalmologic diagnostics tasks

The thesis resolved the important scientific task – increase of the human visual analyzer diagnostics efficiency by implementation of the information technology (IT) analysis of the steady-state visual evoked potentials (SSVEP) based on the new developed mathematical model, methods of statistical estimation and identification of diagnostic parameters, decision-making algorithms. The review determined that most information technologies use an additive model, VEP selection is carried out by averaging of post-stimulus realizations, the decision making is made by comparison of the amplitude-frequency characteristics with the normative values. These facts indicates sources of the improvement in the ophthalmic diagnostics effectiveness through the usage of SSVEP and the consideration of the interconnection between the channels of registration. The brain electrical activity is a result of a large number of excitatory and inhibitory postsynaptic potential generated synapses at random times. Using the theory of linear random processes, the mathematical model in the form of a linear random process is substantiated. Taking into account the SSVEP registration terms the frequency of impulse generation equals the frequency of the external stimulation periodicity, respectively the intensity of impulse emergence will be periodical. Based on contemplations the mathematical model of two-channeled SSVEP is a two-dimensional linear periodical random process which cyclostationarity of mathematical expectation and correlation function is proved. Gaussian Signal Distribution hypothesis is confirmed by the histogram analysis and the normality test using the D'Agostino criteria. The stationary of nested sequences taken over a period is confirmed using the Student's and Fisher's criteria. Whereas the mathematical expectation and correlation function of the two-channeled VEP completely determine the probabilistic distribution of the signal, consequently, they were used as a source of informative characteristics. The orthogonal decomposition based on Chebyshev discrete argument's functions used to identify informative characteristics based on mathematical expectation. To identify the second set of informative characteristics the application of the two-dimensional Karhunen-Loeve decomposition of the two-channel SSVEP’s correlation function is considered as a set of eigenvectors of the correlation function. To implement the classification algorithm, the nearest neighbor method (KNN) was modified by adding the weighting parameters that reflect the importance of the informative characteristics components and using the similarity factor as a metric of the distance between two matrices of eigenvectors. The cross-validation is performed on the input data in order to estimate the KNN parameter values empirically. The client-server architecture solution and appropriate tools were selected for IT implementation. The obtained results of the SSVEP period estimation coincide with the period of external stimulation, which is one of the arguments for confirming the adequacy of the mathematical model. The results of the two-dimensional Karhunen-Loeve decomposition are analyzed. The corresponding number of eigenvectors of the two-channel SSVEP correlation matrix as the second set of diagnostic parameters is determined by analyzing the decomposition components energy contribution to achieve a given energy threshold (95%). Based on the newly formed complex of diagnostic parameters and using the cross-validation, the optimal parameters for the KNN binary classification algorithm were estimated. By using software tools and well-grounded algorithms, the effectiveness of the developed IT of the SSVEP analysis, which allows automated ophthalmic diagnostics in conditions not suitable for standard protocols and takes into account the relationship between the registration channels caused by the biophysical human visual analyzer structure, has been proved.