Objective: The aim of this pilot study was to determine whether automatically extracted speech features, derived from targeted speech tasks, can differentiate Multiple System Atrophy (MSA) from Parkinson’s Disease (PD) and healthy controls (HC) in an English-speaking cohort. The goal is to identify the most informative speech tasks and features for distinguishing between these groups, and to evaluate the potential of the automated system as an accessible and non-invasive diagnostic tool.

Background: Multiple System Atrophy (MSA) is a rapidly progressive neurodegenerative disorder marked by parkinsonian and/or cerebellar features in combination with autonomic dysfunction. A major clinical challenge is the misdiagnosis of MSA as Parkinson’s Disease (PD) due to overlapping symptoms, despite MSA having a more aggressive disease course and poorer response to pharmacological treatment, such as Levodopa. Earlier and accurate discrimination between the disorders is important for prognosis, patient management and access to disease-specific clinical trials.

Method: Speech recordings were collected from participants attending the Dublin Neurological Institute at the Mater Hospital. An automated feature extraction approach was used to derive acoustic, temporal, and lexical variables from several tasks including free speech, picture description, procedural memory and passage reading. The features were then entered into classification models to assess task specific and combined discriminatory performance.

Results: Preliminary analysis of 15 participants (3 MSA, 6 PD, 6 HC) showed promising discriminatory capacity. The best performing combination was for picture description, passage reading and procedural memory, achieving a sensitivity of 0.756, a specificity of 0.875, and a balanced accuracy of 0.756. Across the highest performing models, the most informative features included total pause duration, pause count, dysfluency count and articulation rate.

Conclusion: These preliminary findings suggest that automated speech analysis may provide an accessible and non-invasive approach to support differential diagnosis for MSA. However, further work on larger cohorts of participants is required to establish reproducibility and clinical utility.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/automated-speech-analysis-for-the-differentiation-of-multiple-system-atrophy-from-parkinsons-disease-in-an-english-speaking-cohort/