Vocal Fold Disorders Classification and Optimization of a Custom Video Laryngoscopy Dataset Through Structural...

Area of Science:

• Otolaryngology
• Artificial Intelligence
• Medical Imaging

Background:

• Video laryngoscopy is crucial for diagnosing laryngeal lesions but relies on manual inspection, risking errors and delays.
• Current diagnostic methods can overlook subtle structural changes, impacting early detection and treatment.
• AI offers potential to enhance diagnostic reliability and accelerate the identification of vocal fold abnormalities.

Purpose of the Study:

• To develop and evaluate a hybrid Convolutional Neural Network (CNN) for efficient analysis of video laryngoscopy data.
• To reduce data processing load by eliminating redundant frames and focusing on clinically significant key frames.
• To improve the accuracy and speed of diagnosing laryngeal lesions, including vocal fold nodules and polyps.

Main Methods:

• A hybrid CNN architecture was designed to process video laryngoscopy footage.
• Key frames were extracted using the Structural Similarity Index Measure (SSIM) with varying thresholds.
• The system classified frames from healthy individuals and patients with vocal fold nodules or polyps.

Main Results:

• An SSIM threshold of 0.90 effectively selected informative key frames, balancing data reduction and information content.
• The AI model achieved a 98% overall classification accuracy in identifying laryngeal conditions.
• The approach significantly reduced memory usage and processing time while maintaining high diagnostic performance.

Conclusions:

• The proposed AI-driven method optimizes video laryngoscopy analysis by prioritizing critical information.
• This technique assists physicians in making faster diagnostic decisions for laryngeal pathologies.
• The system effectively reduces computational demands, making advanced analysis more accessible.