The efficacy of acoustic-based articulatory impairment phenotypes for characterizing and classifying divergent neurodegenerative diseases

Introduction Neurodegenerative diseases often lead to communication deficits that can have devastating effects on patient quality of life [1], [2]. Yet, there is no validated set of objective measures that can comprehensively characterize speech abnormalities [3]. Quantitative profiles of neurodegenerative diseases are important for (1) informing differential diagnosis and (2) identifying phenotypic variability that may motivate the development of more targeted treatments. Features from the articulatory subsystem are of particular interest because articulatory deficits are closely related to intelligibility loss [4]–[7] and have demonstrated diagnostic efficacy for different movement disorders [3], [8]–[10]. However, the large number of descriptors used for articulation has created challenges with interpreting measures and deciphering the constructs that may be important for differential diagnosis and treatment [11], [12]. Thus, our recent work sought to develop a unifying framework (i.e., Coordination, Consistency, Speed, Precision, and Rate) for characterizing the diversity of articulatory abnormalities. In the current study, we used this framework to comprehensively profile articulatory impairments in four divergent neurodegenerative diseases: amyotrophic lateral sclerosis (ALS), progressive ataxia (PA), Parkinson’s disease (PD), and nonfluent primary progressive aphasia with progressive apraxia of speech (nfPPA+PAOS). We then assessed the efficacy of articulatory phenotyping for disease classification. Our research questions were: a. What are the articulatory phenotypes of neurodegenerative populations known to have divergent speech motor deficits? b. Do the features contribute unique information to the characterization of articulatory function? a. What is the efficacy of articulatory phenotypes for classifying divergent neurodegenerative diseases? b. Is a profile of articulatory features more diagnostically useful than individual features? Methods Speech samples were obtained from MGH and from several outside collaborators. Acoustic measures were extracted from audio recordings of 217 participants (46 with ALS, 52 with PA, 60 with PD, 20 with nfPPA+PAOS, and 39 controls) during the sequential motion rate task. Acoustic profiles were comprised of the proportion of stop-gap to syllable length (Coordination), across-repetition variability in voice onset time (Consistency), second formant slope in /ka/ (Speed), across-consonant variability in second formant slopes of /pa/, /ta/, /ka/ (Precision), and syllables per second (Rate). Results The primary features that distinguished the groups were reduced Consistency for PA, preserved Rate for PD, and reduced Coordination and Rate for nfPPA+PAOS (see Figure 1). Pairwise correlations revealed weak to moderate correlations for all pairs of acoustic features, suggesting that the features may contribute unique information to the characterization of articulatory function. Univariate and multivariate linear discriminant analyses were then used to determine the classification accuracy for each feature alone and all five features together. ROC curve analyses revealed that disease classification accuracy was strongest when all features were included in the model. Discussion Overall, there was evidence of articulatory impairment phenotypes for the four populations, and the phenotypes demonstrated high classification accuracy for three of the four groups. Our findings highlight the phenotypic variability present across neurodegenerative motor diseases that may (1) provide complementary information for differential diagnosis and (2) inform the development of individualized speech treatments and sensitive outcome measures.