Assessing Laryngeal Neuromotor Activity from Phonation.

Area of Science:

• Neurology
• Speech Science
• Biomedical Engineering

Background:

• Neurodegenerative motor disorders, such as Parkinson's Disease (PD), significantly impact daily life, with speech production being particularly affected.
• Alterations in neuromotor activity (NMA) in laryngeal muscles disrupt phonation stability and regularity, affecting vocal fold control.
• The cricothyroid (tensor) and thyroarytenoid (relaxer) muscle systems, regulated by distinct neuromotor pathways, are crucial for phonation.

Purpose of the Study:

• To investigate the impact of Parkinson's Disease on neuromotor activity (NMA) related to phonation.
• To develop a method for distinguishing between phonations of individuals with Parkinson's Disease (PDPs) and healthy control participants (HCPs) using differential neuromotor activity (DNMA).
• To apply explainable AI (XAI) methods for objective disease grading and stratification.

Main Methods:

• Indirect estimation of muscular tension via inverse filtering to derive differential neuromotor activity (DNMA) in tensor and relaxer pathways.
• Amplitude distributions of DNMA channels were analyzed using Jensen-Shannon distributions to compare PDPs and HCPs against reference mid-age normative participants (RSPs).
• A dataset of 96 phonation samples was used to train decision tree classifiers (DTCs) for distinguishing PDPs from HCPs, with 10-fold cross-validation.

Main Results:

• Decision tree classifiers achieved high accuracy in differentiating male (90.63%) and female (87.50%) PDPs from HCPs.
• Classification performance metrics included high sensitivity (e.g., 93.33% for males) and specificity (e.g., 88.23% for males).
• The study demonstrated the potential for objective disease grading and stratification of PDPs and HCPs using explainable AI.

Conclusions:

• Differential neuromotor activity (DNMA) analysis provides a reliable method for detecting speech alterations in Parkinson's Disease.
• Machine learning classifiers, particularly decision trees, can effectively distinguish individuals with PD from healthy controls based on voice analysis.
• This approach offers a promising tool for objective assessment and stratification of Parkinson's Disease patients, leveraging explainable AI.