Developing an Artificial Intelligence Tool to Predict Vocal Cord Pathology in Primary Care Settings.

Area of Science:

• Medical technology
• Artificial intelligence in healthcare
• Otolaryngology

Background:

• Primary care physicians lack adequate diagnostic tools for voice disorders.
• Voice disorders, or dysphonia, can significantly impact quality of life.
• Early and accurate diagnosis is crucial for effective management of voice pathologies.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) tool for diagnosing voice disorders.
• To assess the AI's ability to differentiate various laryngeal pathologies.
• To determine the AI's utility in assisting primary care physicians with voice disorder diagnosis.

Main Methods:

• Collected prospective voice recordings via smartphones (2019-2021) and utilized the Saarbruecken dataset.
• Converted audio files to mel-spectrograms using TensorFlow for analysis.
• Trained an artificial neural network (ANN) to categorize pathologies including neurological, muscular, inflammatory, and mass lesions.

Main Results:

• The AI achieved a 98% F1-score in detecting abnormal pathology.
• The ANN accurately differentiated specific conditions such as unilateral paralysis, laryngitis, and adductor spasmodic dysphonia (ADSD).
• The AI model outperformed otolaryngologists in a head-to-head comparison on a standardized test set (83% vs. 55%).

Conclusions:

• An AI tool was successfully developed to differentiate vocal cord pathologies with high accuracy.
• Future research should focus on controlled data collection and multi-center collaborations to enhance AI predictive capabilities.
• This AI tool serves as a proof-of-concept to aid primary care physicians in managing patients with dysphonia.