Multi-Objective Optimization of Wavelet-Packet-Based Features in Pathological Diagnosis of Alzheimer Using Spontaneous Speech Signals

Abstract

Alzheimer's disease (AD) ranks among the main types of neurodegenerative disorders. Patients suffering AD should tackle serious problems since their language skills malfunction. The impact of such disorders is reflected by reduced quality and feature variation of spontaneous speech signals in speech analysis. This paper aims at assessing the variations of some specific types of these energy-and entropy-based features within the frequency range of the speech signals. In the approach followed, the wavelet-packet coefficients are utilized to extract the energy and entropy measures at every spectral subband in six successive levels of decomposition. However, the decomposition process conducts a set of high-dimensional feature vectors that is a challenging task for feature selection. This study suggests the application of a Non-dominated Sorting Genetic Algorithm-II (NSGA-II) for enhancing a group of the subband indexes of a wavelet-packet for which the extracted features lead to the highest diagnosis rate of the grouping of Alzheimer's and healthy individuals. The technique proposed here showed that the best overall classification results for both optimized entropy feature vs. energy are more noticeable in discriminating patients with AD from healthy subjects. It is also confirmed the significant impact of multi-objective feature selection on performance of classification (i.e., disease diagnosis) and, its conformity to the disordered nature of the biological signals could help diagnose AD in an efficient manner.

Efficient Pause Extraction and Encode Strategy for Alzheimer’s Disease Detection Using Only Acoustic Features from Spontaneous Speech

“…Eyben et al [ 11 ] proposed a standard acoustic parameter set, the extended Geneva minimalistic acoustic parameter set (eGeMAPS), such that the research results obtained in various areas of automatic speech analysis could be properly compared. Moreover, some researchers explored acoustic features for AD diagnosis from the perspective of digital signal processing, such as higher-order spectral features [ 12 ], fractal features [ 13 ], and wavelet-packet-based features [ 14 ].…”