Ανάπτυξη Και Χρήση Μεθοδολογιών Πολλαπλής Διακριτικής Και Κατευθυντικής Ανάλυσης Βιοϊατρικών Δεδομένων Για Την Ανάδειξη Παθολογικών Προτύπων

Abstract

In the framework of the present PhD thesis, the potential of directional multiscale analysis to detect abnormal patterns in biomedical data is investigated. First, directional multiscale analysis is applied to assess the interaction of neural potentials during a time discrimination psychoacoustic task. Psychoacoustics is the branch of psychophysics which deals with the human perception of the acoustic stimuli (sound), making a sharp distinction between the physical stimulus and psychological response to it. Electroencephalogram (EEG) and Event Related Potential (ERP) signals of twenty five participants in a properly designed psychoacoustics experiment were recorded and analyzed using the Wavelet Coherence and Entropy. According to the results, differences in the pattern of delta, alpha and gamma rhythms are correlated to the Just Noticeable Difference (JND) in pulses duration, calculated by the psychoacoustic analysis. Moreover, the potential of directional multiscale analysis to discriminate symptomatic from asymptomatic carotid artery plaques, from B-mode Ultrasound images was investigated. Texture and Motion characteristics are of great importance for the diagnosis and management of plaque’s instability in carotid atheromatous stenosis. A sample of symptomatic and asymptomatic arteries was interrogated and (directional) multiresolution based texture features were estimated from systolic and diastolic B-mode ultrasound images. In terms of classification accuracy, multiresolution transforms outperformed standard approaches (1st and 2nd order statistics, gray median scale, and fractals), while directional multiresolution outperformed multiresolution transforms. The results demonstrated the superiority of the curvelet transform, in terms of classification accuracy. In addition, the causality plane based on multiresolution analysis was used to assess the complexity of the content of the plaques. Four wavelet entropies and four statistical distances along with 43 values for q paremeter of nonextensive entropies were used. The highest classification rate (91.4%) that outperformed all abovementioned algorithms was achieved by the Tsallis entropy and the Kullback statistical distance for q = 4.2. The findings indicate that asumptomatric plaques exhibit a more ordered (less entropy) and complex (higher statistical complexity) behavior than the symptomatic cases. This fact implies that the evolution of the atheromatous disease trigger a material-organized state. Finally, optimization of multiscale motion estimation parameters of plaques was achieved in terms of the decomposition scheme, the level of analysis and wavelet function used. The optimization is performed in the context of an in silico data framework, consisting of simulated ultrasound image sequences of the carotid artery. SWT, a high-order coiflet function (ex. coif5) and one level of multiscale image decomposition is suggested as the optimal parameterization to achieve maximum accuracy in the particular application.