Acoustic analysis of speech and voice disorders in patients with Lewy Body Diseases

Abstract

Lewy body diseases (LBDs) are a group of neurodegenerative diseases that consists of Parkinson’s disease, Parkinson’s disease with dementia (PD), and dementia with Lewy bodies. LBDs are characterized by the aggregation of 𝛼�-synuclein in specific brain regions, which leads to the formation of Lewy bodies and Lewy neuritis, usually accompanied by neurodegeneration. Identification of early stages in LBDs is crucial since the neurodegeneration may be possibly stopped or treated before the pathological cascades start. This study aims to identify people at high risk of developing LBDs i.e. in prodromal stages of these diseases, based on an acoustic analysis of speech and voice. Eighteen acoustic features were evaluated from 36 Czech native, clinically diagnosed PD patients, 29 healthy controls (HC) and 37 subjects at risk of developing LBDs. Statistical analysis of acoustic features and multiple modelling techniques were explored in order to find the best discriminative model, including multivariate logistic regression and the state-of-the-art algorithm XGBoost. We finally propose an XGBoost model that identifies subjects in high risk of developing LBDs achieving 76 % specificity and 71 % sensitivity on test dataset (including PD and HC). Prosodic and articulatory features were found to be the most discriminant. Thirteen subjects were confirmed to be in high risk of LBDs with our model. Accuracy of these results will be further evaluated in future following a longitudinally monitoring and clinical examination of participants.