|
Fluorescence spectroscopy has been reported as a very promising approach for the discrimination between healthy and atherosclerotic arteries, as far as both, the spectral shape and the intensity of the corresponding spectra seem to be useful parameters for the diagnosis, at specific wavelengths. Nevertheless there are some difficulties in the precise diagnosis, mainly between the different categories of atherosclerotic arteries (fibrous, calcified, heavy calcified). These difficulties are based on the one hand on biophysical factors, such as the necessity for the preknowledge of tissue fluorophores or the complexity of tissue optics. On the other hand, different spectral classification algorithms have been used, such as multivariate linear regression, decision plane analysis and Bayesian decision analysis, each one with certain disadvantages. In this work, two different classification algorithms were developed and evaluated. During the first procedure, simple dimensionless functions were formed by the ratio of the intensities at selected wavelengths and the logistic model was used for statistical analysis. Decision surfaces were drawn and it was estimated that the probability of correct classification is 88%. The algorithm correctly discriminates 97% of healthy from diseased samples and 80% of fibrous from calcified coronary arteries. During the second procedure, a proper ratio was selected in the sense that the ratio groups of the populations P1 and P2 might be separated with an essential considerable veracity probability. The separability was confirmed by testing the validity of specific statistical hypotheses. The demonstration has been made by means of the Kolmogorov-Smirnov goodness of fit method. Therefore by applying statistical methods on proper parameters obtained from the specimens spectra, it has been able to automatically classify the arterial specimens into healthy (normal), fibrous, calcified and heavily calcified, with more than 99.9% probability (less than 0.1% confidence interval). The different classifications algorithms are thoroughly discussed and evaluated.
|
