Early motor signs of Parkinson's disease detected by acoustic speech analysis and classification methods

Purpose: Parkinson's disease (PD) is a slowly progressing and highly debilitating disease. By the time it is diagnosed there is already substantial damage to the central nervous system. There is no medical treatment yet to prevent or decelerate the disease process. Brain imaging and other technological methods can detect the disease earlier than by clinical examination, but such technology is extremely expensive. Speech abnormalities might be among the earlier manifestations of the disease. However, they might be too subtle to be detectable perceptually. Acoustic analysis of speech is objective, valid and inexpensive method. The purpose of this study was to find predictors of early motor signs of Parkinson's disease (EMSPD) by acoustic speech analysis and classification methods. Methods: Twenty seven individuals with EMSPD (mean age= 63.56+10.50; H&Y=1.59±0.42; UPDRS (motor)= 19.07±8.38; years since diagnosis= 1.48±0.51), all optimally medicated during the study, and 86 healthy, age-matched, controls participated in the study. They sustained vowel phonation and read a paragraph. Potential predictors of PD risk were age, gender, and acoustic measures of vowels, voice fundamental frequency, temporal aspects speech articulation, and measures of vocal stability. Results: A multivariate stepwise selection model process yielded four surviving predictors, all reflecting vocal and articulatory instability. ROC area under the curve (AUC) was 0.905. At logistic regression probability 38% or higher, sensitivity was 78.8%, specificity 88.1%, with overall 85.5% correct prediction. Conclusions: Detection of EMSPD by speech acoustic analysis and classification methods is feasible. Whether these methods can detect speech abnormalities in the prodromal/preclinical stage is yet to be explored.