Voiceprint and machine learning models for early detection of bulbar dysfunction in ALS

Area of Science:

• Neurology
• Speech Pathology
• Biomedical Engineering

Background:

• Bulbar dysfunction, a motor neuron disorder in ALS, affects speech and swallowing.
• Early symptoms include voice deterioration, impacting articulation and speech quality.
• Current diagnostic methods may not detect subtle, early-stage changes.

Purpose of the Study:

• To develop an automated methodology for early diagnosis of bulbar dysfunction using voice analysis.
• To create a voiceprint analysis tool that surpasses current clinical detection capabilities.
• To enable earlier intervention and disease management for ALS patients.

Main Methods:

• A novel voiceprint was created from sustained Spanish vowels.
• Principal component analysis was used to extract key vocal features.
• Supervised and semi-supervised machine learning models (Random Forest, SVM) were employed for classification.

Main Results:

• The Random Forest model achieved 88.3% accuracy in distinguishing bulbar from control participants.
• After semi-supervised relabeling, SVM achieved 91.0% accuracy, 83.3% sensitivity, and 100.0% specificity for bulbar vs. non-bulbar classification.
• The proposed method demonstrated superior diagnostic performance over existing clinical and computational approaches.

Conclusions:

• The developed voice analysis methodology is effective and applicable for early bulbar dysfunction detection.
• This approach can lead to an affordable and user-friendly tool for early diagnosis and disease monitoring.
• Earlier and more accurate diagnosis can significantly benefit ALS patient care and treatment strategies.