Through the Speech and Vocal Signals Hidden Secrets: An Explainable Methodology for Neurological Diseases Early...

Area of Science:

• Computational Neuroscience
• Biomedical Engineering
• Speech Pathology

Background:

• Neurodegenerative diseases cause progressive neurological damage, impacting function.
• Early diagnosis is crucial for effective treatment and improved patient quality of life.
• Current diagnostic methods for neurodegenerative diseases often lack non-invasive, scalable solutions.

Purpose of the Study:

• To develop and validate a non-invasive methodology for early neurodegenerative disease diagnosis using voice analysis.
• To identify reliable speech and vocal biomarkers for distinguishing between healthy and pathological voices.
• To assess the performance of machine learning algorithms in classifying voice signals.

Main Methods:

• Extraction of acoustic, articulation, and cepstral features from vowel and speech signals.
• Application of machine learning algorithms trained on a combined dataset of voice features.
• Validation of the methodology on a dataset comprising normophonic and pathological voice samples.

Main Results:

• The proposed methodology achieved high accuracy (97.5%) in distinguishing healthy from pathological voices.
• Excellent performance metrics were reported, including sensitivity (98.5%), precision (97.0%), F1-score (98.0%), MCC (0.95), and AUC (0.98).
• Explainability tasks confirmed the reliability of identified neurological biomarkers from speech and vocal features.

Conclusions:

• Voice analysis, combined with machine learning, presents a promising non-invasive tool for early neurodegenerative disease screening.
• The developed methodology demonstrates high reliability and accuracy for clinical applications.
• Further research and clinical validation can lead to approved, large-scale diagnostic tools for neurodegenerative conditions.