Indices from visibility graph complexity of spontaneous speech signal: An efficient nonlinear tool for Alzheimer's disease diagnosis

Interest in dynamical data analytics of human speech related to the diagnosis of Alzheimer's disease (AD) has recently risen. This study aims to scrutinize the dynamic variations in AD speech based on concepts called complexity and fractality. Towards this goal, the visibility graph (VG) of the spontaneous speech time series as a quantitative approach is proposed to distinguish healthy subjects from those with Alzheimer's. The dynamical patterns of the speech signals are analyzed between three stages of AD and healthy subjects by examining some specific kinds of these measures, known as graph index complexity (GIC) and power of scale-freeness (PS) in VG. The results show the practical ability of the GIC and PS to detect the underlying pathological mechanism of Alzheimer's disease. In addition, it is found that all speech series have visibility graphs with a power-law topology, and fractality in them is reflected by a mechanism related to the brain system pathology, which affects the language skills of people. •Complex network theory is applied to study nonlinear characteristics of Alzheimer's speech.•Differential complexities between healthy and Alzheimer's subjects are quantified using complexity and fractality.•The statistical properties of the network with visibility graph quantification are investigated.•The elicited performance of the visibility graph of AD speech is remarkable.