Application of Automated Data Analytics and Machine Learning to Develop Explainable Objective Markers for Differential Assessment of Progressive Motor Speech Disorders in Neurodegenerative Diseases

Background: Oral diadochokinetic (DDK) tasks are commonly used in clinical and research settings to assess neuromotor speech disorders. Due to their simple phonological structure, these tasks are feasible for individuals across a wide range of functional capacities and relatively resistant to cognitive-linguistic deficits, which frequently co-occur with motor impairments in multisystem disorders such as neurodegenerative diseases (NDDs). Moreover, the rhythmicity of these tasks facilitates automated or semi-automated analysis using computerized algorithms. Prior study has demonstrated the feasibility of this algorithmic approach by implementing fine-grained temporal analyses of multimodal oral DDK performance for presymptomatic detection of motor speech impairments in amyotrophic lateral sclerosis (ALS). Building on these prior efforts, the goal of this study is to adapt and apply the fine-grained multimodal temporal DDK analyses to another NDD population—Parkinson’s disease (PD)—to identify disease-specific oral DDK impairments and to evaluate their utility for differential diagnosis of motor speech disorders in PD and ALS. Methods: Craniofacial myoelectric, kinematic, and acoustic data were collected from 16 individuals with ALS, 13 individuals with PD, and 10 neurologically healthy controls (HC), during three oral DDK tasks (i.e., rapid repetitions of [pɑ], [tɑ], or [kɑ]). Using a semiautomated data-analytic pipeline, twenty-six temporal features were extracted across the three modalities and subsequently clustered into a lower-dimensional set of composite measures via factor analysis. The features were compared across the three groups (ALS, PD, HC) using linear mixed-effects models. The composite measures were analyzed using (1) supervised machine learning classification methods to evaluate their discriminatory performance for distinguishing among ALS, PD, and HC groups, and (2) general linear models to examine their relationships with two standardized functional metrics—communication efficiency and Montreal Cognitive Assessment (MoCA) scores. Results: The DDK features were clustered into three interpretable composite measures reflecting temporal variability, duration, and synchrony, respectively. Significant group effects were found for the majority of features, revealing robust disease-specific impairment patterns characterized by prolonged duration in ALS and increased temporal variability in PD. All machine learning models demonstrated satisfactory performance, each achieving an area under the curve (AUC) greater than 0.80 for all pairwise classifications. The composite measures (1) exhibited differential associations with communication efficiency across groups, and (2) no correlation with MoCA scores. Conclusions: This study presented a potential methodological pipeline that integrated knowledge-based and data-driven methods to derive three physiologically interpretable and clinically meaningful objective markers that captured subclinical changes in oral DDK performance in both ALS and PD. These markers demonstrated strong discriminatory performance in detecting and differentiating disease-specific impairment patterns in ALS and PD and were associated with functional speech decline, while remaining minimally influenced by concurrent cognitive-linguistic impairment. These findings support the potential of automation-based data analytics and machine learning techniques to complement conventional clinical methods and facilitate early detection and differential diagnosis of progressive motor speech disorders, ultimately advancing timely, individualized care in neurodegenerative diseases.