A pattern recognition approach to spasmodic dysphonia and muscle tension dysphonia automatic classification

Area of Science:

• Laryngology
• Computational Linguistics
• Artificial Intelligence

Background:

• Spasmodic dysphonia (SD) and muscle tension dysphonia (MTD) are distinct voice disorders with overlapping symptoms, complicating accurate diagnosis.
• Misdiagnosis can lead to inappropriate treatment, as SD often requires interventions like botulinum toxin injections, while MTD is typically managed with voice therapy.

Purpose of the Study:

• To develop and evaluate machine learning models, specifically neural networks and support vector machines, for differentiating between normal, SD, and MTD voices.
• To compare the efficacy of these computational methods against previous classification techniques for voice disorders.

Main Methods:

• Extraction of eight acoustic parameters from sustained vowel /a/ sounds.
• Application of pattern recognition algorithms, including neural networks and support vector machines, for voice classification.
• Comparative analysis of classification performance for normal vs. pathological (SD and MTD) voices and for distinguishing between SD and MTD.

Main Results:

• The proposed methods achieved superior classification rates for distinguishing normal from pathological voices compared to previous studies.
• The algorithms demonstrated high effectiveness in accurately differentiating between muscle tension dysphonia and spasmodic dysphonia.

Conclusions:

• Computational methods utilizing neural networks and support vector machines offer a promising tool to aid clinicians in the differential diagnosis of SD and MTD.
• Accurate AI-assisted diagnosis is crucial for selecting the most effective therapeutic strategies for patients with these voice disorders.