Modelagem incremental fuzzy para detecção incipiente e estimação do grau de severidade da doença de parkinson a partir de sinais de voz

Abstract

Parkinson’s disease is a chronic neurodegenerative disorder that affects the central nervous system and therefore the motor system. Many early non-motor symptoms are in principle hard to be perceived by the individual. As the disease develops, symptoms become noticeable. Currently, motor impairment is essential to support the clinical diagnosis. Among the research directions on detecting the Parkinson’s disease in early stage – prior to motor symptoms – is that of monitoring the voice of individuals and subtle changes during speech. Frequency spectrum analysis may reveal the disease in early stage. The present study considers methods and models for Incremental Machine Learning from the Computational Intelligence perspective. The Fuzzy Set-based Evolving Model (FBeM) for detecting patterns of the Parkinson’s disease from sustained phonation is a nonlinear and nonstationary model, that is, it is able to self-adapt over time from a data stream. Experimental data were obtained from the University of Oxford Parkinson’s Voice Initiative. The data are related to attributes of the frequency spectrum of 42 individuals, being 23 on early stage Parkinson’s disease. The developed models provide an estimation of the severity of the disease according to the Unified Parkinson Disease Rating Scale (UDPRS). A neuro-fuzzy modeling approach, known as Adaptive Neuro-Fuzzy Inference System (ANFIS), is considered for comparisons. Moreover, linear and monotic correlations were analysed for attribute selection. Estimation results have shown that the performance of the proposed evolving FBeM model slightly overcomes that of ANFIS.