Emprego do estimador KSG na classificação de patologias laríngeas por intermédio da análise dinâmica não-linear.

Abstract

Voice disorders and laryngeal pathologies may affect different age groups. The human being brings many of these deviations with him/her from birth, or might still develop throughout life. Due to its non-stationarity and its chaotic dynamics, voice signals can not be correctly analyzed from traditional methods of analyzing experimental signals. Thus, the caos theory, an area of non-linear dynamical systems theory, applied in non-linear time series, has been adopted as a new non-linear approach to voice signal processing. In this context, this thesis proposes to investigate the applicability of a estimator of mutual information (KSG estimator) and their variants (KOLE estimator), based on the nearest k-neighbors, in the estimation of measurements of non linear dynamic analysis: reconstruction delay estimation of phase space in dynamical systems (τ ), first minimum of the mutual information function (PMIM), Shannon’s entropy (H) and correlation entropy(K2). Studies have revealed that the KSG estimator is less biased than the naive estimator commonly used, which motivated us to apply it to detect the presence of voice disorders caused by laryngeal pathologies such as Reinke’s edema, paralysis on the vocal folds and nodules on the vocal folds. Two types of classifiers were used to obtain results: one based on discriminant analysis and one based on a support vector approach (SVM). The comparison of the results obtained with the naive estimator and the KSG estimator, for both classifiers used, shows that the KSG estimator had an improved performance in terms of accuracy, sensitivity and specificity, thus revealing an efficient estimation method.